diff --git a/README.md b/README.md index 2ec23c5..3b20f7d 100644 --- a/README.md +++ b/README.md @@ -34,7 +34,8 @@ After that, every PR merge re-deploys automatically; `workflow_dispatch` re-runs best-effort benchmarks (honest verified/unverified) - **Hardware-fit boxes** — NVIDIA (8–96 GB incl. RTX 5090 32 GB), AMD (8–192 GB incl. RX 7900 XT 20 GB), Intel Arc (SYCL/Vulkan), Snapdragon X (unified memory), - MacBook / Mac Studio (Metal), DGX, Android, iPhone — recomputed live from the + MacBook / Mac Mini / Mac Studio / Mac Pro (Metal), DGX, Android, iPhone — + recomputed live from the selected quant (est VRAM + 1.5 GB headroom vs tier), with backend labels (CUDA / ROCm / Metal / SYCL+Vulkan) on every section - **Run locally with → recommended server** — for every (model × hardware) pair the diff --git a/index.html b/index.html index 9438d1d..3247178 100644 --- a/index.html +++ b/index.html @@ -96,6 +96,8 @@

what-llm Snapdragon X (unified) + + @@ -127,6 +129,8 @@

what-llm Snapdragon X (unified) + + @@ -134,6 +138,7 @@

what-llm +
@@ -162,6 +167,7 @@

What is it good for?

+
@@ -217,6 +223,9 @@

Side-by-side

{k:"64gb",v:64},{k:"96gb",v:96},{k:"128gb",v:128}], mac_studio: [{k:"32gb",v:32},{k:"64gb",v:64},{k:"96gb",v:96},{k:"128gb",v:128}, {k:"192gb",v:192},{k:"256gb",v:256},{k:"512gb",v:512}], + mac_mini: [{k:"16gb",v:16},{k:"24gb",v:24},{k:"32gb",v:32},{k:"48gb",v:48}, + {k:"64gb",v:64},{k:"128gb",v:128}], + mac_pro: [{k:"192gb",v:192},{k:"384gb",v:384},{k:"512gb",v:512}], dgx: [{k:"640gb",v:640},{k:"1128gb",v:1128},{k:"1440gb",v:1440}], android:[{k:"8gb",v:8},{k:"12gb",v:12},{k:"16gb",v:16},{k:"24gb",v:24}], iphone: [{k:"8gb",v:8},{k:"12gb",v:12}] @@ -228,12 +237,20 @@

Side-by-side

{sec:"snapdragon", label:"Snapdragon X — CPU / Adreno Vulkan (unified)"}, {sec:"macbook", label:"MacBook — Metal (unified memory)"}, {sec:"mac_studio", label:"Mac Studio — Metal (unified memory)"}, + {sec:"mac_mini", label:"Mac Mini — Metal (unified memory)"}, + {sec:"mac_pro", label:"Mac Pro — Metal (unified memory)"}, {sec:"dgx", label:"NVIDIA DGX — CUDA (total GPU memory)"}, {sec:"android", label:"Android (CPU)"}, {sec:"iphone", label:"iPhone (CPU)"} ]; - var MAC_CATS = {macbook: true, mac_studio: true, snapdragon: true}; + var MAC_CATS = {macbook: true, mac_studio: true, mac_mini: true, mac_pro: true, snapdragon: true}; function catLabel(c){ var f = SECTIONS.filter(function(x){ return x.sec === c; }); return f.length ? f[0].label : c; } + // agentic coding = coding/reasoning/agentic model AND long context (>= 32K) + function agenticOk(m){ + var tags = (m.best_for || []).join(" ").toLowerCase(); + if(!/(agentic|coding|reasoning|code)/.test(tags)) return false; + return (m.context_window || 0) >= 32768; + } function esc(t){ return String(t == null ? "" : t).replace(/[&<>"']/g, function(ch){ return {"&":"&","<":"<",">":">","\"":""","'":"'"}[ch]; }); } // inference-server catalog (inference-server-catalog, issue #27) — mirrors // src/whatllm/servers.py; backends are booleans per hardware category @@ -588,6 +605,13 @@

Side-by-side

function renderProfile(m){ var prof = m.profile || {}; $("#d-summary").textContent = prof.summary || ""; + var ag = $("#d-agentic"); + if(agenticOk(m)){ + ag.style.display = ""; + ag.innerHTML = "🤖 Agentic coding ready — coding/reasoning model with " + (m.context_window || 0).toLocaleString() + " token context"; + } else { + ag.style.display = "none"; + } var bf = $("#d-bestfor"); bf.textContent = ""; (prof.best_for || []).forEach(function(uc){ var s = document.createElement("span"); @@ -686,6 +710,7 @@

Side-by-side

var picks = state.models.filter(function(m){ if(needUse && (m.best_for || []).indexOf(use) === -1) return false; if($("#w-commercial").checked && m.commercial_ok === false) return false; + if($("#w-agentic").checked && !agenticOk(m)) return false; return fitsFor(m, cat, tier); }); picks.sort(function(a,b){ return b.trending_score - a.trending_score; }); @@ -702,7 +727,7 @@

Side-by-side

row.appendChild(a); var why = document.createElement("div"); why.className = "why"; why.textContent = (m.best_for || []).slice(0,3).join(" · ") + - " · ~" + m.est_vram_gb + " GB est."; + " · ~" + m.est_vram_gb + " GB est." + (agenticOk(m) ? " · 🤖 agentic coding" : ""); row.appendChild(why); var recs = recommendServers(m, cat); if(recs.length){ diff --git a/models/Qwen__Qwen2.5-14B-Instruct.json b/models/Qwen__Qwen2.5-14B-Instruct.json index 96efb2a..26657f3 100644 --- a/models/Qwen__Qwen2.5-14B-Instruct.json +++ b/models/Qwen__Qwen2.5-14B-Instruct.json @@ -69,6 +69,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -78,6 +86,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/Qwen__Qwen2.5-3B-Instruct.json b/models/Qwen__Qwen2.5-3B-Instruct.json index 335e5a3..f60c627 100644 --- a/models/Qwen__Qwen2.5-3B-Instruct.json +++ b/models/Qwen__Qwen2.5-3B-Instruct.json @@ -75,6 +75,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -84,6 +92,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/Qwen__Qwen2.5-72B-Instruct.json b/models/Qwen__Qwen2.5-72B-Instruct.json index 8fa982e..3b99271 100644 --- a/models/Qwen__Qwen2.5-72B-Instruct.json +++ b/models/Qwen__Qwen2.5-72B-Instruct.json @@ -69,6 +69,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": false, + "24gb": false, + "32gb": false, + "48gb": false, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": false, "64gb": true, @@ -78,6 +86,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/Qwen__Qwen3-32B.json b/models/Qwen__Qwen3-32B.json index a052614..a2ad448 100644 --- a/models/Qwen__Qwen3-32B.json +++ b/models/Qwen__Qwen3-32B.json @@ -69,6 +69,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": false, + "24gb": false, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -78,6 +86,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/Qwen__Qwen3-8B.json b/models/Qwen__Qwen3-8B.json index f8b4403..85ef2ec 100644 --- a/models/Qwen__Qwen3-8B.json +++ b/models/Qwen__Qwen3-8B.json @@ -75,6 +75,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -84,6 +92,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/bartowski__Meta-Llama-3.1-8B-Instruct-GGUF.json b/models/bartowski__Meta-Llama-3.1-8B-Instruct-GGUF.json index 905b518..3de81ac 100644 --- a/models/bartowski__Meta-Llama-3.1-8B-Instruct-GGUF.json +++ b/models/bartowski__Meta-Llama-3.1-8B-Instruct-GGUF.json @@ -81,6 +81,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -90,6 +98,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/bundle.js b/models/bundle.js index cd590a0..bf819da 100644 --- a/models/bundle.js +++ b/models/bundle.js @@ -1 +1 @@ -window.MODELS_BUNDLE = {"microsoft/Phi-3-mini-4k-instruct":{"id":"microsoft/Phi-3-mini-4k-instruct","name":"Phi-3 Mini 4K Instruct","author":"microsoft","parameters_b":3.82,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/microsoft/Phi-3-mini-4k-instruct","trending_score":4120,"downloads":1450000,"quants":[{"name":"Q4_K_M","size_gb":2.34,"estimated_vram_gb":3.64,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":2.72,"estimated_vram_gb":4.02,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":4.06,"estimated_vram_gb":5.36,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":7.64,"estimated_vram_gb":8.94,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~3.64 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~3.64 GB est.) on flagship iPhone"}},"profile":{"summary":"Phi: compact models ideal for edge/mobile and constrained hardware.","best_for":["chat","reasoning","mobile"],"strengths":["Supports up to 4,096 tokens of context","Small footprint (3-4B) fits phones and low-end GPUs","Surprisingly strong reasoning for size"],"weaknesses":["Smaller knowledge base","License terms vary by version (MIT/other)"],"limitations":["Short context window (4,096 tokens)"],"provenance":[{"claim":"Supports up to 4,096 tokens of context","source":"config","confidence":"high"},{"claim":"Short context window (4,096 tokens)","source":"config","confidence":"medium"},{"claim":"Small footprint (3-4B) fits phones and low-end GPUs","source":"curated","confidence":"medium"},{"claim":"Surprisingly strong reasoning for size","source":"curated","confidence":"medium"},{"claim":"Smaller knowledge base","source":"curated","confidence":"medium"},{"claim":"License terms vary by version (MIT/other)","source":"curated","confidence":"medium"}]},"license":"mit","commercial_ok":true,"context_window":4096,"model_type":"instruct","languages":["en"],"knowledge_cutoff":"2023-10","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen3-8B":{"id":"Qwen/Qwen3-8B","name":"Qwen3 8B","author":"Qwen","parameters_b":8.19,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen3-8B","trending_score":3890,"downloads":3200000,"quants":[{"name":"Q4_K_M","size_gb":5.01,"estimated_vram_gb":6.31,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.84,"estimated_vram_gb":7.14,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":8.71,"estimated_vram_gb":10.01,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":16.38,"estimated_vram_gb":17.68,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag","agentic"],"strengths":["Supports up to 40,960 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 40,960 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"apache-2.0","commercial_ok":true,"context_window":40960,"model_type":"chat","languages":["en","zh"],"knowledge_cutoff":"2025-03","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"meta-llama/Llama-3.1-8B-Instruct":{"id":"meta-llama/Llama-3.1-8B-Instruct","name":"Llama 3.1 8B Instruct","author":"meta-llama","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct","trending_score":2870,"downloads":9800000,"quants":[{"name":"Q4_K_M","size_gb":4.91,"estimated_vram_gb":6.21,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.73,"estimated_vram_gb":7.03,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":8.54,"estimated_vram_gb":9.84,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":16.06,"estimated_vram_gb":17.36,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.1","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"bartowski/Meta-Llama-3.1-8B-Instruct-GGUF":{"id":"bartowski/Meta-Llama-3.1-8B-Instruct-GGUF","name":"Llama 3.1 8B Instruct GGUF","author":"bartowski","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF","trending_score":1980,"downloads":278000,"quants":[{"name":"Q4_K_S","size_gb":4.69,"estimated_vram_gb":5.99,"notes":"Slightly lower quality with more space savings"},{"name":"Q4_K_M","size_gb":4.92,"estimated_vram_gb":6.22,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.73,"estimated_vram_gb":7.03,"notes":"Higher quality, slightly larger"},{"name":"Q6_K","size_gb":6.6,"estimated_vram_gb":7.9,"notes":"Very high quality"},{"name":"Q8_0","size_gb":8.54,"estimated_vram_gb":9.84,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.1","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen2.5-14B-Instruct":{"id":"Qwen/Qwen2.5-14B-Instruct","name":"Qwen2.5 14B Instruct","author":"Qwen","parameters_b":14.77,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen2.5-14B-Instruct","trending_score":1560,"downloads":5100000,"quants":[{"name":"Q4_K_M","size_gb":9.04,"estimated_vram_gb":10.34,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":10.53,"estimated_vram_gb":11.83,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":15.7,"estimated_vram_gb":17.0,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag"],"strengths":["Supports up to 32,768 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"qwen","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","zh"],"knowledge_cutoff":"2024-09","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"mistralai/Mixtral-8x7B-Instruct-v0.1":{"id":"mistralai/Mixtral-8x7B-Instruct-v0.1","name":"Mixtral 8x7B Instruct","author":"mistralai","parameters_b":46.7,"architecture":"moe","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1","trending_score":1120,"downloads":4300000,"quants":[{"name":"Q4_K_M","size_gb":28.58,"estimated_vram_gb":29.88,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":33.3,"estimated_vram_gb":34.6,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":49.64,"estimated_vram_gb":50.94,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":false,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Mixtral: sparse MoE — near-larger-model quality at lower active-parameter cost.","best_for":["chat","reasoning"],"strengths":["Supports up to 32,768 tokens of context","MoE efficiency: strong quality per active parameter","Good multilingual support"],"weaknesses":["High TOTAL parameter count — needs lots of VRAM/RAM (all experts resident)"],"limitations":["32GB-class machines are borderline at Q4 (see hardware flags)"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"MoE efficiency: strong quality per active parameter","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"High TOTAL parameter count — needs lots of VRAM/RAM (all experts resident)","source":"curated","confidence":"medium"},{"claim":"32GB-class machines are borderline at Q4 (see hardware flags)","source":"curated","confidence":"medium"}]},"license":"apache-2.0","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","fr","de","es","it"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen2.5-72B-Instruct":{"id":"Qwen/Qwen2.5-72B-Instruct","name":"Qwen2.5 72B Instruct","author":"Qwen","parameters_b":72.71,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct","trending_score":940,"downloads":2700000,"quants":[{"name":"Q4_K_M","size_gb":44.5,"estimated_vram_gb":45.8,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":51.84,"estimated_vram_gb":53.14,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":77.29,"estimated_vram_gb":78.59,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":false,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag"],"strengths":["Supports up to 131,072 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"qwen","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","zh"],"knowledge_cutoff":"2024-09","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"deepseek-ai/DeepSeek-R1":{"id":"deepseek-ai/DeepSeek-R1","name":"DeepSeek R1","author":"deepseek-ai","parameters_b":671.0,"architecture":"moe","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/deepseek-ai/DeepSeek-R1","trending_score":860,"downloads":6500000,"quants":[{"name":"Q4_K_M","size_gb":410.65,"estimated_vram_gb":411.95,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":478.42,"estimated_vram_gb":479.72,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":713.27,"estimated_vram_gb":714.57,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"96gb":false},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"192gb":false},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":false},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":false,"96gb":false,"128gb":false},"mac_studio":{"32gb":false,"64gb":false,"96gb":false,"128gb":false,"192gb":false,"256gb":false,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"DeepSeek: frontier-scale reasoning and coding MoE models.","best_for":["reasoning","coding","chat"],"strengths":["Supports up to 131,072 tokens of context","Frontier reasoning and math","MoE design keeps active params low"],"weaknesses":["Very large total parameter count — datacenter class","Specialized tooling for some variants"],"limitations":["Consumer GPUs rarely fit even the smallest quants","Gated on some variants"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Frontier reasoning and math","source":"curated","confidence":"medium"},{"claim":"MoE design keeps active params low","source":"curated","confidence":"medium"},{"claim":"Very large total parameter count — datacenter class","source":"curated","confidence":"medium"},{"claim":"Specialized tooling for some variants","source":"curated","confidence":"medium"},{"claim":"Consumer GPUs rarely fit even the smallest quants","source":"curated","confidence":"medium"},{"claim":"Gated on some variants","source":"curated","confidence":"medium"}]},"license":"deepseek","commercial_ok":true,"context_window":131072,"model_type":"reasoner","languages":["en","zh"],"knowledge_cutoff":"2024-11","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"meta-llama/Llama-3.2-1B":{"id":"meta-llama/Llama-3.2-1B","name":"Llama 3.2 1B","author":"meta-llama","parameters_b":1.23,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/meta-llama/Llama-3.2-1B","trending_score":640,"downloads":2100000,"quants":[{"name":"Q4_K_M","size_gb":0.75,"estimated_vram_gb":2.05,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":0.88,"estimated_vram_gb":2.18,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":1.31,"estimated_vram_gb":2.61,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":2.46,"estimated_vram_gb":3.76,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~2.05 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~2.05 GB est.) on flagship iPhone"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Base (pre-instruct) model — use the instruct/chat variant for conversation","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Base (pre-instruct) model — use the instruct/chat variant for conversation","source":"curated","confidence":"high"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.2","commercial_ok":true,"context_window":131072,"model_type":"base","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen2.5-3B-Instruct":{"id":"Qwen/Qwen2.5-3B-Instruct","name":"Qwen2.5 3B Instruct","author":"Qwen","parameters_b":3.09,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen2.5-3B-Instruct","trending_score":720,"downloads":1800000,"quants":[{"name":"Q4_K_M","size_gb":1.89,"estimated_vram_gb":3.19,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":2.2,"estimated_vram_gb":3.5,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":3.28,"estimated_vram_gb":4.58,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":6.18,"estimated_vram_gb":7.48,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~3.19 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~3.19 GB est.) on flagship iPhone"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag"],"strengths":["Supports up to 32,768 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"qwen","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","zh"],"knowledge_cutoff":"2024-09","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"google/gemma-3-4b-it":{"id":"google/gemma-3-4b-it","name":"Gemma 3 4B IT","author":"google","parameters_b":3.8,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/google/gemma-3-4b-it","trending_score":590,"downloads":890000,"quants":[{"name":"Q4_K_M","size_gb":2.33,"estimated_vram_gb":3.63,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":2.71,"estimated_vram_gb":4.01,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":4.04,"estimated_vram_gb":5.34,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":7.6,"estimated_vram_gb":8.9,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~3.63 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~3.63 GB est.) on flagship iPhone"}},"profile":{"summary":"Gemma: open Google models, strong at general chat; multilingual on newer versions.","best_for":["chat","reasoning"],"strengths":["Supports up to 32,768 tokens of context","Good general chat quality","Multilingual in 3.x versions"],"weaknesses":["Gated card on some versions","Smaller ecosystem than Llama/Qwen"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Google license terms apply"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Good general chat quality","source":"curated","confidence":"medium"},{"claim":"Multilingual in 3.x versions","source":"curated","confidence":"medium"},{"claim":"Gated card on some versions","source":"curated","confidence":"medium"},{"claim":"Smaller ecosystem than Llama/Qwen","source":"curated","confidence":"medium"},{"claim":"Google license terms apply","source":"curated","confidence":"medium"}]},"license":"gemma","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","es","de","fr","ja","ko","pt","zh"],"knowledge_cutoff":"2025-02","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen3-32B":{"id":"Qwen/Qwen3-32B","name":"Qwen3 32B","author":"Qwen","parameters_b":32.76,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen3-32B","trending_score":830,"downloads":410000,"quants":[{"name":"Q4_K_M","size_gb":20.05,"estimated_vram_gb":21.35,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":23.36,"estimated_vram_gb":24.66,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":34.82,"estimated_vram_gb":36.12,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":true,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"apache-2.0","commercial_ok":true,"context_window":131072,"model_type":"chat","languages":["en","zh"],"knowledge_cutoff":"2025-03","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"meta-llama/Llama-3.1-70B-Instruct":{"id":"meta-llama/Llama-3.1-70B-Instruct","name":"Llama 3.1 70B Instruct","author":"meta-llama","parameters_b":70.6,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct","trending_score":710,"downloads":1500000,"quants":[{"name":"Q4_K_M","size_gb":43.21,"estimated_vram_gb":44.51,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":50.34,"estimated_vram_gb":51.64,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":75.05,"estimated_vram_gb":76.35,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":false,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.1","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"unsloth/DeepSeek-V4-Flash-0731-GGUF":{"id":"unsloth/DeepSeek-V4-Flash-0731-GGUF","name":"DeepSeek V4 Flash 0731 GGUF","author":"unsloth","parameters_b":304.18,"architecture":"moe","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/unsloth/DeepSeek-V4-Flash-0731-GGUF","trending_score":625,"downloads":189000,"quants":[{"name":"UD-IQ1_M","size_gb":86.9,"estimated_vram_gb":88.2,"notes":"Aggressive 1-bit dynamic quant — only practical option for 128 GB Macs"},{"name":"UD-Q2_K_XL","size_gb":96.8,"estimated_vram_gb":98.1,"notes":"2-bit XL quant, compact"},{"name":"UD-Q3_K_XL","size_gb":128.2,"estimated_vram_gb":129.5,"notes":"3-bit XL quant"},{"name":"UD-Q4_K_XL","size_gb":155.1,"estimated_vram_gb":156.4,"notes":"4-bit XL quant — higher quality"},{"name":"UD-Q8_K_XL","size_gb":161.9,"estimated_vram_gb":163.2,"notes":"8-bit XL quant — highest quality"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":false},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":false,"96gb":true,"128gb":true},"mac_studio":{"32gb":false,"64gb":false,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"DeepSeek: frontier-scale reasoning and coding MoE models.","best_for":["reasoning","coding"],"strengths":["Supports up to 131,072 tokens of context","Frontier reasoning and math","MoE design keeps active params low"],"weaknesses":["Very large total parameter count — datacenter class","Specialized tooling for some variants"],"limitations":["Consumer GPUs rarely fit even the smallest quants","Gated on some variants"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Frontier reasoning and math","source":"curated","confidence":"medium"},{"claim":"MoE design keeps active params low","source":"curated","confidence":"medium"},{"claim":"Very large total parameter count — datacenter class","source":"curated","confidence":"medium"},{"claim":"Specialized tooling for some variants","source":"curated","confidence":"medium"},{"claim":"Consumer GPUs rarely fit even the smallest quants","source":"curated","confidence":"medium"},{"claim":"Gated on some variants","source":"curated","confidence":"medium"}]},"license":"deepseek","commercial_ok":true,"context_window":131072,"model_type":"reasoner","languages":["en","zh"],"knowledge_cutoff":"2025-07","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"cognitivecomputations/dolphin-2.9-llama-3.1-8b":{"id":"cognitivecomputations/dolphin-2.9-llama-3.1-8b","name":"Dolphin 2.9 Llama 3.1 8B","author":"cognitivecomputations","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/cognitivecomputations/dolphin-2.9-llama-3.1-8b","trending_score":540,"downloads":920000,"quants":[{"name":"Q4_K_M","size_gb":4.91,"estimated_vram_gb":6.21,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.73,"estimated_vram_gb":7.03,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":8.54,"estimated_vram_gb":9.84,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":16.06,"estimated_vram_gb":17.36,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Non-commercial license (cc-by-nc-4.0) — check before shipping","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Non-commercial license (cc-by-nc-4.0) — check before shipping","source":"tags","confidence":"high"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"cc-by-nc-4.0","commercial_ok":false,"context_window":131072,"model_type":"chat","languages":["en"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"}}; +window.MODELS_BUNDLE = {"microsoft/Phi-3-mini-4k-instruct":{"id":"microsoft/Phi-3-mini-4k-instruct","name":"Phi-3 Mini 4K Instruct","author":"microsoft","parameters_b":3.82,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/microsoft/Phi-3-mini-4k-instruct","trending_score":4120,"downloads":1450000,"quants":[{"name":"Q4_K_M","size_gb":2.34,"estimated_vram_gb":3.64,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":2.72,"estimated_vram_gb":4.02,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":4.06,"estimated_vram_gb":5.36,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":7.64,"estimated_vram_gb":8.94,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~3.64 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~3.64 GB est.) on flagship iPhone"}},"profile":{"summary":"Phi: compact models ideal for edge/mobile and constrained hardware.","best_for":["chat","reasoning","mobile"],"strengths":["Supports up to 4,096 tokens of context","Small footprint (3-4B) fits phones and low-end GPUs","Surprisingly strong reasoning for size"],"weaknesses":["Smaller knowledge base","License terms vary by version (MIT/other)"],"limitations":["Short context window (4,096 tokens)"],"provenance":[{"claim":"Supports up to 4,096 tokens of context","source":"config","confidence":"high"},{"claim":"Short context window (4,096 tokens)","source":"config","confidence":"medium"},{"claim":"Small footprint (3-4B) fits phones and low-end GPUs","source":"curated","confidence":"medium"},{"claim":"Surprisingly strong reasoning for size","source":"curated","confidence":"medium"},{"claim":"Smaller knowledge base","source":"curated","confidence":"medium"},{"claim":"License terms vary by version (MIT/other)","source":"curated","confidence":"medium"}]},"license":"mit","commercial_ok":true,"context_window":4096,"model_type":"instruct","languages":["en"],"knowledge_cutoff":"2023-10","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen3-8B":{"id":"Qwen/Qwen3-8B","name":"Qwen3 8B","author":"Qwen","parameters_b":8.19,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen3-8B","trending_score":3890,"downloads":3200000,"quants":[{"name":"Q4_K_M","size_gb":5.01,"estimated_vram_gb":6.31,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.84,"estimated_vram_gb":7.14,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":8.71,"estimated_vram_gb":10.01,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":16.38,"estimated_vram_gb":17.68,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag","agentic"],"strengths":["Supports up to 40,960 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 40,960 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"apache-2.0","commercial_ok":true,"context_window":40960,"model_type":"chat","languages":["en","zh"],"knowledge_cutoff":"2025-03","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"meta-llama/Llama-3.1-8B-Instruct":{"id":"meta-llama/Llama-3.1-8B-Instruct","name":"Llama 3.1 8B Instruct","author":"meta-llama","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct","trending_score":2870,"downloads":9800000,"quants":[{"name":"Q4_K_M","size_gb":4.91,"estimated_vram_gb":6.21,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.73,"estimated_vram_gb":7.03,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":8.54,"estimated_vram_gb":9.84,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":16.06,"estimated_vram_gb":17.36,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.1","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"bartowski/Meta-Llama-3.1-8B-Instruct-GGUF":{"id":"bartowski/Meta-Llama-3.1-8B-Instruct-GGUF","name":"Llama 3.1 8B Instruct GGUF","author":"bartowski","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF","trending_score":1980,"downloads":278000,"quants":[{"name":"Q4_K_S","size_gb":4.69,"estimated_vram_gb":5.99,"notes":"Slightly lower quality with more space savings"},{"name":"Q4_K_M","size_gb":4.92,"estimated_vram_gb":6.22,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.73,"estimated_vram_gb":7.03,"notes":"Higher quality, slightly larger"},{"name":"Q6_K","size_gb":6.6,"estimated_vram_gb":7.9,"notes":"Very high quality"},{"name":"Q8_0","size_gb":8.54,"estimated_vram_gb":9.84,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.1","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen2.5-14B-Instruct":{"id":"Qwen/Qwen2.5-14B-Instruct","name":"Qwen2.5 14B Instruct","author":"Qwen","parameters_b":14.77,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen2.5-14B-Instruct","trending_score":1560,"downloads":5100000,"quants":[{"name":"Q4_K_M","size_gb":9.04,"estimated_vram_gb":10.34,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":10.53,"estimated_vram_gb":11.83,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":15.7,"estimated_vram_gb":17.0,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag"],"strengths":["Supports up to 32,768 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"qwen","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","zh"],"knowledge_cutoff":"2024-09","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"mistralai/Mixtral-8x7B-Instruct-v0.1":{"id":"mistralai/Mixtral-8x7B-Instruct-v0.1","name":"Mixtral 8x7B Instruct","author":"mistralai","parameters_b":46.7,"architecture":"moe","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1","trending_score":1120,"downloads":4300000,"quants":[{"name":"Q4_K_M","size_gb":28.58,"estimated_vram_gb":29.88,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":33.3,"estimated_vram_gb":34.6,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":49.64,"estimated_vram_gb":50.94,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":false,"24gb":false,"32gb":false,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":false,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Mixtral: sparse MoE — near-larger-model quality at lower active-parameter cost.","best_for":["chat","reasoning"],"strengths":["Supports up to 32,768 tokens of context","MoE efficiency: strong quality per active parameter","Good multilingual support"],"weaknesses":["High TOTAL parameter count — needs lots of VRAM/RAM (all experts resident)"],"limitations":["32GB-class machines are borderline at Q4 (see hardware flags)"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"MoE efficiency: strong quality per active parameter","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"High TOTAL parameter count — needs lots of VRAM/RAM (all experts resident)","source":"curated","confidence":"medium"},{"claim":"32GB-class machines are borderline at Q4 (see hardware flags)","source":"curated","confidence":"medium"}]},"license":"apache-2.0","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","fr","de","es","it"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen2.5-72B-Instruct":{"id":"Qwen/Qwen2.5-72B-Instruct","name":"Qwen2.5 72B Instruct","author":"Qwen","parameters_b":72.71,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct","trending_score":940,"downloads":2700000,"quants":[{"name":"Q4_K_M","size_gb":44.5,"estimated_vram_gb":45.8,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":51.84,"estimated_vram_gb":53.14,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":77.29,"estimated_vram_gb":78.59,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":true,"128gb":true},"mac_studio":{"32gb":false,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag"],"strengths":["Supports up to 131,072 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"qwen","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","zh"],"knowledge_cutoff":"2024-09","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"deepseek-ai/DeepSeek-R1":{"id":"deepseek-ai/DeepSeek-R1","name":"DeepSeek R1","author":"deepseek-ai","parameters_b":671.0,"architecture":"moe","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/deepseek-ai/DeepSeek-R1","trending_score":860,"downloads":6500000,"quants":[{"name":"Q4_K_M","size_gb":410.65,"estimated_vram_gb":411.95,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":478.42,"estimated_vram_gb":479.72,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":713.27,"estimated_vram_gb":714.57,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"96gb":false},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"192gb":false},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":false},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":false,"96gb":false,"128gb":false},"mac_mini":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":false,"128gb":false},"mac_studio":{"32gb":false,"64gb":false,"96gb":false,"128gb":false,"192gb":false,"256gb":false,"512gb":true},"mac_pro":{"192gb":false,"384gb":false,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"DeepSeek: frontier-scale reasoning and coding MoE models.","best_for":["reasoning","coding","chat"],"strengths":["Supports up to 131,072 tokens of context","Frontier reasoning and math","MoE design keeps active params low"],"weaknesses":["Very large total parameter count — datacenter class","Specialized tooling for some variants"],"limitations":["Consumer GPUs rarely fit even the smallest quants","Gated on some variants"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Frontier reasoning and math","source":"curated","confidence":"medium"},{"claim":"MoE design keeps active params low","source":"curated","confidence":"medium"},{"claim":"Very large total parameter count — datacenter class","source":"curated","confidence":"medium"},{"claim":"Specialized tooling for some variants","source":"curated","confidence":"medium"},{"claim":"Consumer GPUs rarely fit even the smallest quants","source":"curated","confidence":"medium"},{"claim":"Gated on some variants","source":"curated","confidence":"medium"}]},"license":"deepseek","commercial_ok":true,"context_window":131072,"model_type":"reasoner","languages":["en","zh"],"knowledge_cutoff":"2024-11","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"meta-llama/Llama-3.2-1B":{"id":"meta-llama/Llama-3.2-1B","name":"Llama 3.2 1B","author":"meta-llama","parameters_b":1.23,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/meta-llama/Llama-3.2-1B","trending_score":640,"downloads":2100000,"quants":[{"name":"Q4_K_M","size_gb":0.75,"estimated_vram_gb":2.05,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":0.88,"estimated_vram_gb":2.18,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":1.31,"estimated_vram_gb":2.61,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":2.46,"estimated_vram_gb":3.76,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~2.05 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~2.05 GB est.) on flagship iPhone"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Base (pre-instruct) model — use the instruct/chat variant for conversation","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Base (pre-instruct) model — use the instruct/chat variant for conversation","source":"curated","confidence":"high"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.2","commercial_ok":true,"context_window":131072,"model_type":"base","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen2.5-3B-Instruct":{"id":"Qwen/Qwen2.5-3B-Instruct","name":"Qwen2.5 3B Instruct","author":"Qwen","parameters_b":3.09,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen2.5-3B-Instruct","trending_score":720,"downloads":1800000,"quants":[{"name":"Q4_K_M","size_gb":1.89,"estimated_vram_gb":3.19,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":2.2,"estimated_vram_gb":3.5,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":3.28,"estimated_vram_gb":4.58,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":6.18,"estimated_vram_gb":7.48,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~3.19 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~3.19 GB est.) on flagship iPhone"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag"],"strengths":["Supports up to 32,768 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"qwen","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","zh"],"knowledge_cutoff":"2024-09","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"google/gemma-3-4b-it":{"id":"google/gemma-3-4b-it","name":"Gemma 3 4B IT","author":"google","parameters_b":3.8,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/google/gemma-3-4b-it","trending_score":590,"downloads":890000,"quants":[{"name":"Q4_K_M","size_gb":2.33,"estimated_vram_gb":3.63,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":2.71,"estimated_vram_gb":4.01,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":4.04,"estimated_vram_gb":5.34,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":7.6,"estimated_vram_gb":8.9,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":true,"12gb":true,"16gb":true,"24gb":true,"note":"Practical at the smallest quant (~3.63 GB est.) on flagship Android"},"iphone":{"8gb":true,"12gb":true,"note":"Practical at the smallest quant (~3.63 GB est.) on flagship iPhone"}},"profile":{"summary":"Gemma: open Google models, strong at general chat; multilingual on newer versions.","best_for":["chat","reasoning"],"strengths":["Supports up to 32,768 tokens of context","Good general chat quality","Multilingual in 3.x versions"],"weaknesses":["Gated card on some versions","Smaller ecosystem than Llama/Qwen"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Google license terms apply"],"provenance":[{"claim":"Supports up to 32,768 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Good general chat quality","source":"curated","confidence":"medium"},{"claim":"Multilingual in 3.x versions","source":"curated","confidence":"medium"},{"claim":"Gated card on some versions","source":"curated","confidence":"medium"},{"claim":"Smaller ecosystem than Llama/Qwen","source":"curated","confidence":"medium"},{"claim":"Google license terms apply","source":"curated","confidence":"medium"}]},"license":"gemma","commercial_ok":true,"context_window":32768,"model_type":"instruct","languages":["en","es","de","fr","ja","ko","pt","zh"],"knowledge_cutoff":"2025-02","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"Qwen/Qwen3-32B":{"id":"Qwen/Qwen3-32B","name":"Qwen3 32B","author":"Qwen","parameters_b":32.76,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/Qwen/Qwen3-32B","trending_score":830,"downloads":410000,"quants":[{"name":"Q4_K_M","size_gb":20.05,"estimated_vram_gb":21.35,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":23.36,"estimated_vram_gb":24.66,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":34.82,"estimated_vram_gb":36.12,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":true,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":false,"24gb":false,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Qwen: strong general-purpose chat, reasoning and coding; long context; multilingual.","best_for":["chat","reasoning","coding","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Strong instruction following and general chat","Good multilingual support","Long context windows (often 32k+)"],"weaknesses":["Heavier than comparable 7-8B models for the same quality","Some variants need specific tooling for best results"],"limitations":["Knowledge cutoff varies by variant (see model card)","Coder/reasoner variants are specialized"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Strong instruction following and general chat","source":"curated","confidence":"medium"},{"claim":"Good multilingual support","source":"curated","confidence":"medium"},{"claim":"Long context windows (often 32k+)","source":"curated","confidence":"medium"},{"claim":"Heavier than comparable 7-8B models for the same quality","source":"curated","confidence":"medium"},{"claim":"Some variants need specific tooling for best results","source":"curated","confidence":"medium"},{"claim":"Knowledge cutoff varies by variant (see model card)","source":"curated","confidence":"medium"},{"claim":"Coder/reasoner variants are specialized","source":"curated","confidence":"medium"}]},"license":"apache-2.0","commercial_ok":true,"context_window":131072,"model_type":"chat","languages":["en","zh"],"knowledge_cutoff":"2025-03","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"meta-llama/Llama-3.1-70B-Instruct":{"id":"meta-llama/Llama-3.1-70B-Instruct","name":"Llama 3.1 70B Instruct","author":"meta-llama","parameters_b":70.6,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct","trending_score":710,"downloads":1500000,"quants":[{"name":"Q4_K_M","size_gb":43.21,"estimated_vram_gb":44.51,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":50.34,"estimated_vram_gb":51.64,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":75.05,"estimated_vram_gb":76.35,"notes":"Near-lossless, needs more VRAM"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":true,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":true},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":true,"128gb":true},"mac_studio":{"32gb":false,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Model card is gated — README not accessible to anonymous crawls","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Model card is gated — README not accessible to anonymous crawls","source":"config","confidence":"medium"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"llama3.1","commercial_ok":true,"context_window":131072,"model_type":"instruct","languages":["en","de","fr","es","hi","pt","zh"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"unsloth/DeepSeek-V4-Flash-0731-GGUF":{"id":"unsloth/DeepSeek-V4-Flash-0731-GGUF","name":"DeepSeek V4 Flash 0731 GGUF","author":"unsloth","parameters_b":304.18,"architecture":"moe","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/unsloth/DeepSeek-V4-Flash-0731-GGUF","trending_score":625,"downloads":189000,"quants":[{"name":"UD-IQ1_M","size_gb":86.9,"estimated_vram_gb":88.2,"notes":"Aggressive 1-bit dynamic quant — only practical option for 128 GB Macs"},{"name":"UD-Q2_K_XL","size_gb":96.8,"estimated_vram_gb":98.1,"notes":"2-bit XL quant, compact"},{"name":"UD-Q3_K_XL","size_gb":128.2,"estimated_vram_gb":129.5,"notes":"3-bit XL quant"},{"name":"UD-Q4_K_XL","size_gb":155.1,"estimated_vram_gb":156.4,"notes":"4-bit XL quant — higher quality"},{"name":"UD-Q8_K_XL","size_gb":161.9,"estimated_vram_gb":163.2,"notes":"8-bit XL quant — highest quality"}],"hardware":{"nvidia":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"96gb":true},"amd":{"8gb":false,"12gb":false,"16gb":false,"20gb":false,"24gb":false,"32gb":false,"48gb":false,"192gb":true},"intel_arc":{"8gb":false,"10gb":false,"12gb":false,"16gb":false},"snapdragon":{"16gb":false,"32gb":false,"64gb":false},"macbook":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":false,"96gb":true,"128gb":true},"mac_mini":{"16gb":false,"24gb":false,"32gb":false,"48gb":false,"64gb":false,"128gb":true},"mac_studio":{"32gb":false,"64gb":false,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"DeepSeek: frontier-scale reasoning and coding MoE models.","best_for":["reasoning","coding"],"strengths":["Supports up to 131,072 tokens of context","Frontier reasoning and math","MoE design keeps active params low"],"weaknesses":["Very large total parameter count — datacenter class","Specialized tooling for some variants"],"limitations":["Consumer GPUs rarely fit even the smallest quants","Gated on some variants"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Frontier reasoning and math","source":"curated","confidence":"medium"},{"claim":"MoE design keeps active params low","source":"curated","confidence":"medium"},{"claim":"Very large total parameter count — datacenter class","source":"curated","confidence":"medium"},{"claim":"Specialized tooling for some variants","source":"curated","confidence":"medium"},{"claim":"Consumer GPUs rarely fit even the smallest quants","source":"curated","confidence":"medium"},{"claim":"Gated on some variants","source":"curated","confidence":"medium"}]},"license":"deepseek","commercial_ok":true,"context_window":131072,"model_type":"reasoner","languages":["en","zh"],"knowledge_cutoff":"2025-07","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"},"cognitivecomputations/dolphin-2.9-llama-3.1-8b":{"id":"cognitivecomputations/dolphin-2.9-llama-3.1-8b","name":"Dolphin 2.9 Llama 3.1 8B","author":"cognitivecomputations","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","hf_url":"https://huggingface.co/cognitivecomputations/dolphin-2.9-llama-3.1-8b","trending_score":540,"downloads":920000,"quants":[{"name":"Q4_K_M","size_gb":4.91,"estimated_vram_gb":6.21,"notes":"Recommended balanced quant"},{"name":"Q5_K_M","size_gb":5.73,"estimated_vram_gb":7.03,"notes":"Higher quality, slightly larger"},{"name":"Q8_0","size_gb":8.54,"estimated_vram_gb":9.84,"notes":"Near-lossless, needs more VRAM"},{"name":"FP16","size_gb":16.06,"estimated_vram_gb":17.36,"notes":"Full precision, most VRAM"}],"hardware":{"nvidia":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"96gb":true},"amd":{"8gb":true,"12gb":true,"16gb":true,"20gb":true,"24gb":true,"32gb":true,"48gb":true,"192gb":true},"intel_arc":{"8gb":true,"10gb":true,"12gb":true,"16gb":true},"snapdragon":{"16gb":true,"32gb":true,"64gb":true},"macbook":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"96gb":true,"128gb":true},"mac_mini":{"16gb":true,"24gb":true,"32gb":true,"48gb":true,"64gb":true,"128gb":true},"mac_studio":{"32gb":true,"64gb":true,"96gb":true,"128gb":true,"192gb":true,"256gb":true,"512gb":true},"mac_pro":{"192gb":true,"384gb":true,"512gb":true},"dgx":{"640gb":true,"1128gb":true,"1440gb":true},"android":{"8gb":false,"12gb":false,"16gb":false,"24gb":false,"note":"Too large for Android phones"},"iphone":{"8gb":false,"12gb":false,"note":"Too large for iPhone phones"}},"profile":{"summary":"Llama: solid general-purpose chat with broad ecosystem support (llama.cpp, vLLM).","best_for":["chat","rag","agentic"],"strengths":["Supports up to 131,072 tokens of context","Excellent ecosystem and quantization support","Well-tuned instruct variants"],"weaknesses":["Benchmarks often matched by newer smaller models","Gated card — README not anonymous-accessible"],"limitations":["Non-commercial license (cc-by-nc-4.0) — check before shipping","Llama community license: commercial use ok but MAU >700M needs Meta approval","Gated model card (login required for files)"],"provenance":[{"claim":"Supports up to 131,072 tokens of context","source":"config","confidence":"high"},{"claim":"Non-commercial license (cc-by-nc-4.0) — check before shipping","source":"tags","confidence":"high"},{"claim":"Excellent ecosystem and quantization support","source":"curated","confidence":"medium"},{"claim":"Well-tuned instruct variants","source":"curated","confidence":"medium"},{"claim":"Benchmarks often matched by newer smaller models","source":"curated","confidence":"medium"},{"claim":"Gated card — README not anonymous-accessible","source":"curated","confidence":"medium"},{"claim":"Llama community license: commercial use ok but MAU >700M needs Meta approval","source":"curated","confidence":"medium"},{"claim":"Gated model card (login required for files)","source":"curated","confidence":"medium"}]},"license":"cc-by-nc-4.0","commercial_ok":false,"context_window":131072,"model_type":"chat","languages":["en"],"knowledge_cutoff":"2023-12","benchmarks":[],"servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"last_updated":"2026-08-10"}}; diff --git a/models/cognitivecomputations__dolphin-2.9-llama-3.1-8b.json b/models/cognitivecomputations__dolphin-2.9-llama-3.1-8b.json index d0a1282..6d7fa68 100644 --- a/models/cognitivecomputations__dolphin-2.9-llama-3.1-8b.json +++ b/models/cognitivecomputations__dolphin-2.9-llama-3.1-8b.json @@ -75,6 +75,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -84,6 +92,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/deepseek-ai__DeepSeek-R1.json b/models/deepseek-ai__DeepSeek-R1.json index a3e7972..746803c 100644 --- a/models/deepseek-ai__DeepSeek-R1.json +++ b/models/deepseek-ai__DeepSeek-R1.json @@ -69,6 +69,14 @@ "96gb": false, "128gb": false }, + "mac_mini": { + "16gb": false, + "24gb": false, + "32gb": false, + "48gb": false, + "64gb": false, + "128gb": false + }, "mac_studio": { "32gb": false, "64gb": false, @@ -78,6 +86,11 @@ "256gb": false, "512gb": true }, + "mac_pro": { + "192gb": false, + "384gb": false, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/google__gemma-3-4b-it.json b/models/google__gemma-3-4b-it.json index acb1742..8387ee9 100644 --- a/models/google__gemma-3-4b-it.json +++ b/models/google__gemma-3-4b-it.json @@ -75,6 +75,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -84,6 +92,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/index.js b/models/index.js index 89b6421..e4c9e22 100644 --- a/models/index.js +++ b/models/index.js @@ -1,2 +1,2 @@ window.MODELS_META = {"crawled_at":"2026-08-10","count":15}; -window.MODELS_INDEX = [{"id":"microsoft/Phi-3-mini-4k-instruct","name":"Phi-3 Mini 4K Instruct","author":"microsoft","parameters_b":3.82,"architecture":"dense","pipeline_tag":"text-generation","trending_score":4120,"downloads":1450000,"model_type":"instruct","commercial_ok":true,"license":"mit","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":3.64,"best_for":["chat","reasoning","mobile"]},{"id":"Qwen/Qwen3-8B","name":"Qwen3 8B","author":"Qwen","parameters_b":8.19,"architecture":"dense","pipeline_tag":"text-generation","trending_score":3890,"downloads":3200000,"model_type":"chat","commercial_ok":true,"license":"apache-2.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":6.31,"best_for":["chat","reasoning","coding","rag","agentic"]},{"id":"meta-llama/Llama-3.1-8B-Instruct","name":"Llama 3.1 8B Instruct","author":"meta-llama","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","trending_score":2870,"downloads":9800000,"model_type":"instruct","commercial_ok":true,"license":"llama3.1","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":6.21,"best_for":["chat","rag","agentic"]},{"id":"bartowski/Meta-Llama-3.1-8B-Instruct-GGUF","name":"Llama 3.1 8B Instruct GGUF","author":"bartowski","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","trending_score":1980,"downloads":278000,"model_type":"instruct","commercial_ok":true,"license":"llama3.1","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":5.99,"best_for":["chat","rag","agentic"]},{"id":"Qwen/Qwen2.5-14B-Instruct","name":"Qwen2.5 14B Instruct","author":"Qwen","parameters_b":14.77,"architecture":"dense","pipeline_tag":"text-generation","trending_score":1560,"downloads":5100000,"model_type":"instruct","commercial_ok":true,"license":"qwen","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":10.34,"best_for":["chat","reasoning","coding","rag"]},{"id":"mistralai/Mixtral-8x7B-Instruct-v0.1","name":"Mixtral 8x7B Instruct","author":"mistralai","parameters_b":46.7,"architecture":"moe","pipeline_tag":"text-generation","trending_score":1120,"downloads":4300000,"model_type":"instruct","commercial_ok":true,"license":"apache-2.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":29.88,"best_for":["chat","reasoning"]},{"id":"Qwen/Qwen2.5-72B-Instruct","name":"Qwen2.5 72B Instruct","author":"Qwen","parameters_b":72.71,"architecture":"dense","pipeline_tag":"text-generation","trending_score":940,"downloads":2700000,"model_type":"instruct","commercial_ok":true,"license":"qwen","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":45.8,"best_for":["chat","reasoning","coding","rag"]},{"id":"deepseek-ai/DeepSeek-R1","name":"DeepSeek R1","author":"deepseek-ai","parameters_b":671.0,"architecture":"moe","pipeline_tag":"text-generation","trending_score":860,"downloads":6500000,"model_type":"reasoner","commercial_ok":true,"license":"deepseek","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":411.95,"best_for":["reasoning","coding","chat"]},{"id":"meta-llama/Llama-3.2-1B","name":"Llama 3.2 1B","author":"meta-llama","parameters_b":1.23,"architecture":"dense","pipeline_tag":"text-generation","trending_score":640,"downloads":2100000,"model_type":"base","commercial_ok":true,"license":"llama3.2","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":2.05,"best_for":["chat","rag","agentic"]},{"id":"Qwen/Qwen2.5-3B-Instruct","name":"Qwen2.5 3B Instruct","author":"Qwen","parameters_b":3.09,"architecture":"dense","pipeline_tag":"text-generation","trending_score":720,"downloads":1800000,"model_type":"instruct","commercial_ok":true,"license":"qwen","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":3.19,"best_for":["chat","reasoning","coding","rag"]},{"id":"google/gemma-3-4b-it","name":"Gemma 3 4B IT","author":"google","parameters_b":3.8,"architecture":"dense","pipeline_tag":"text-generation","trending_score":590,"downloads":890000,"model_type":"instruct","commercial_ok":true,"license":"gemma","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":3.63,"best_for":["chat","reasoning"]},{"id":"Qwen/Qwen3-32B","name":"Qwen3 32B","author":"Qwen","parameters_b":32.76,"architecture":"dense","pipeline_tag":"text-generation","trending_score":830,"downloads":410000,"model_type":"chat","commercial_ok":true,"license":"apache-2.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":21.35,"best_for":["chat","reasoning","coding","rag","agentic"]},{"id":"meta-llama/Llama-3.1-70B-Instruct","name":"Llama 3.1 70B Instruct","author":"meta-llama","parameters_b":70.6,"architecture":"dense","pipeline_tag":"text-generation","trending_score":710,"downloads":1500000,"model_type":"instruct","commercial_ok":true,"license":"llama3.1","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":44.51,"best_for":["chat","rag","agentic"]},{"id":"unsloth/DeepSeek-V4-Flash-0731-GGUF","name":"DeepSeek V4 Flash 0731 GGUF","author":"unsloth","parameters_b":304.18,"architecture":"moe","pipeline_tag":"text-generation","trending_score":625,"downloads":189000,"model_type":"reasoner","commercial_ok":true,"license":"deepseek","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":88.2,"best_for":["reasoning","coding"]},{"id":"cognitivecomputations/dolphin-2.9-llama-3.1-8b","name":"Dolphin 2.9 Llama 3.1 8B","author":"cognitivecomputations","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","trending_score":540,"downloads":920000,"model_type":"chat","commercial_ok":false,"license":"cc-by-nc-4.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"est_vram_gb":6.21,"best_for":["chat","rag","agentic"]}]; +window.MODELS_INDEX = [{"id":"microsoft/Phi-3-mini-4k-instruct","name":"Phi-3 Mini 4K Instruct","author":"microsoft","parameters_b":3.82,"architecture":"dense","pipeline_tag":"text-generation","trending_score":4120,"downloads":1450000,"model_type":"instruct","commercial_ok":true,"license":"mit","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":4096,"est_vram_gb":3.64,"best_for":["chat","reasoning","mobile"]},{"id":"Qwen/Qwen3-8B","name":"Qwen3 8B","author":"Qwen","parameters_b":8.19,"architecture":"dense","pipeline_tag":"text-generation","trending_score":3890,"downloads":3200000,"model_type":"chat","commercial_ok":true,"license":"apache-2.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":40960,"est_vram_gb":6.31,"best_for":["chat","reasoning","coding","rag","agentic"]},{"id":"meta-llama/Llama-3.1-8B-Instruct","name":"Llama 3.1 8B Instruct","author":"meta-llama","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","trending_score":2870,"downloads":9800000,"model_type":"instruct","commercial_ok":true,"license":"llama3.1","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":6.21,"best_for":["chat","rag","agentic"]},{"id":"bartowski/Meta-Llama-3.1-8B-Instruct-GGUF","name":"Llama 3.1 8B Instruct GGUF","author":"bartowski","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","trending_score":1980,"downloads":278000,"model_type":"instruct","commercial_ok":true,"license":"llama3.1","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":5.99,"best_for":["chat","rag","agentic"]},{"id":"Qwen/Qwen2.5-14B-Instruct","name":"Qwen2.5 14B Instruct","author":"Qwen","parameters_b":14.77,"architecture":"dense","pipeline_tag":"text-generation","trending_score":1560,"downloads":5100000,"model_type":"instruct","commercial_ok":true,"license":"qwen","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":32768,"est_vram_gb":10.34,"best_for":["chat","reasoning","coding","rag"]},{"id":"mistralai/Mixtral-8x7B-Instruct-v0.1","name":"Mixtral 8x7B Instruct","author":"mistralai","parameters_b":46.7,"architecture":"moe","pipeline_tag":"text-generation","trending_score":1120,"downloads":4300000,"model_type":"instruct","commercial_ok":true,"license":"apache-2.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":32768,"est_vram_gb":29.88,"best_for":["chat","reasoning"]},{"id":"Qwen/Qwen2.5-72B-Instruct","name":"Qwen2.5 72B Instruct","author":"Qwen","parameters_b":72.71,"architecture":"dense","pipeline_tag":"text-generation","trending_score":940,"downloads":2700000,"model_type":"instruct","commercial_ok":true,"license":"qwen","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":45.8,"best_for":["chat","reasoning","coding","rag"]},{"id":"deepseek-ai/DeepSeek-R1","name":"DeepSeek R1","author":"deepseek-ai","parameters_b":671.0,"architecture":"moe","pipeline_tag":"text-generation","trending_score":860,"downloads":6500000,"model_type":"reasoner","commercial_ok":true,"license":"deepseek","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":411.95,"best_for":["reasoning","coding","chat"]},{"id":"meta-llama/Llama-3.2-1B","name":"Llama 3.2 1B","author":"meta-llama","parameters_b":1.23,"architecture":"dense","pipeline_tag":"text-generation","trending_score":640,"downloads":2100000,"model_type":"base","commercial_ok":true,"license":"llama3.2","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":2.05,"best_for":["chat","rag","agentic"]},{"id":"Qwen/Qwen2.5-3B-Instruct","name":"Qwen2.5 3B Instruct","author":"Qwen","parameters_b":3.09,"architecture":"dense","pipeline_tag":"text-generation","trending_score":720,"downloads":1800000,"model_type":"instruct","commercial_ok":true,"license":"qwen","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":32768,"est_vram_gb":3.19,"best_for":["chat","reasoning","coding","rag"]},{"id":"google/gemma-3-4b-it","name":"Gemma 3 4B IT","author":"google","parameters_b":3.8,"architecture":"dense","pipeline_tag":"text-generation","trending_score":590,"downloads":890000,"model_type":"instruct","commercial_ok":true,"license":"gemma","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":32768,"est_vram_gb":3.63,"best_for":["chat","reasoning"]},{"id":"Qwen/Qwen3-32B","name":"Qwen3 32B","author":"Qwen","parameters_b":32.76,"architecture":"dense","pipeline_tag":"text-generation","trending_score":830,"downloads":410000,"model_type":"chat","commercial_ok":true,"license":"apache-2.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":21.35,"best_for":["chat","reasoning","coding","rag","agentic"]},{"id":"meta-llama/Llama-3.1-70B-Instruct","name":"Llama 3.1 70B Instruct","author":"meta-llama","parameters_b":70.6,"architecture":"dense","pipeline_tag":"text-generation","trending_score":710,"downloads":1500000,"model_type":"instruct","commercial_ok":true,"license":"llama3.1","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":44.51,"best_for":["chat","rag","agentic"]},{"id":"unsloth/DeepSeek-V4-Flash-0731-GGUF","name":"DeepSeek V4 Flash 0731 GGUF","author":"unsloth","parameters_b":304.18,"architecture":"moe","pipeline_tag":"text-generation","trending_score":625,"downloads":189000,"model_type":"reasoner","commercial_ok":true,"license":"deepseek","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":88.2,"best_for":["reasoning","coding"]},{"id":"cognitivecomputations/dolphin-2.9-llama-3.1-8b","name":"Dolphin 2.9 Llama 3.1 8B","author":"cognitivecomputations","parameters_b":8.03,"architecture":"dense","pipeline_tag":"text-generation","trending_score":540,"downloads":920000,"model_type":"chat","commercial_ok":false,"license":"cc-by-nc-4.0","servers":["llama.cpp","llama-server","Ollama","LM Studio","koboldcpp","Jan","oobabooga","llamafile","gpt4all"],"context_window":131072,"est_vram_gb":6.21,"best_for":["chat","rag","agentic"]}]; diff --git a/models/index.json b/models/index.json index b870b06..2474137 100644 --- a/models/index.json +++ b/models/index.json @@ -22,6 +22,7 @@ "llamafile", "gpt4all" ], + "context_window": 4096, "est_vram_gb": 3.64, "best_for": [ "chat", @@ -52,6 +53,7 @@ "llamafile", "gpt4all" ], + "context_window": 40960, "est_vram_gb": 6.31, "best_for": [ "chat", @@ -84,6 +86,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 6.21, "best_for": [ "chat", @@ -114,6 +117,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 5.99, "best_for": [ "chat", @@ -144,6 +148,7 @@ "llamafile", "gpt4all" ], + "context_window": 32768, "est_vram_gb": 10.34, "best_for": [ "chat", @@ -175,6 +180,7 @@ "llamafile", "gpt4all" ], + "context_window": 32768, "est_vram_gb": 29.88, "best_for": [ "chat", @@ -204,6 +210,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 45.8, "best_for": [ "chat", @@ -235,6 +242,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 411.95, "best_for": [ "reasoning", @@ -265,6 +273,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 2.05, "best_for": [ "chat", @@ -295,6 +304,7 @@ "llamafile", "gpt4all" ], + "context_window": 32768, "est_vram_gb": 3.19, "best_for": [ "chat", @@ -326,6 +336,7 @@ "llamafile", "gpt4all" ], + "context_window": 32768, "est_vram_gb": 3.63, "best_for": [ "chat", @@ -355,6 +366,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 21.35, "best_for": [ "chat", @@ -387,6 +399,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 44.51, "best_for": [ "chat", @@ -417,6 +430,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 88.2, "best_for": [ "reasoning", @@ -446,6 +460,7 @@ "llamafile", "gpt4all" ], + "context_window": 131072, "est_vram_gb": 6.21, "best_for": [ "chat", diff --git a/models/meta-llama__Llama-3.1-70B-Instruct.json b/models/meta-llama__Llama-3.1-70B-Instruct.json index 1fb6f08..1215247 100644 --- a/models/meta-llama__Llama-3.1-70B-Instruct.json +++ b/models/meta-llama__Llama-3.1-70B-Instruct.json @@ -69,6 +69,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": false, + "24gb": false, + "32gb": false, + "48gb": false, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": false, "64gb": true, @@ -78,6 +86,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/meta-llama__Llama-3.1-8B-Instruct.json b/models/meta-llama__Llama-3.1-8B-Instruct.json index 71aeb06..9eb59f3 100644 --- a/models/meta-llama__Llama-3.1-8B-Instruct.json +++ b/models/meta-llama__Llama-3.1-8B-Instruct.json @@ -75,6 +75,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -84,6 +92,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/meta-llama__Llama-3.2-1B.json b/models/meta-llama__Llama-3.2-1B.json index fae03d4..80e7e77 100644 --- a/models/meta-llama__Llama-3.2-1B.json +++ b/models/meta-llama__Llama-3.2-1B.json @@ -75,6 +75,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -84,6 +92,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/microsoft__Phi-3-mini-4k-instruct.json b/models/microsoft__Phi-3-mini-4k-instruct.json index 736974d..ab19a8a 100644 --- a/models/microsoft__Phi-3-mini-4k-instruct.json +++ b/models/microsoft__Phi-3-mini-4k-instruct.json @@ -75,6 +75,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": true, + "24gb": true, + "32gb": true, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": true, "64gb": true, @@ -84,6 +92,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/mistralai__Mixtral-8x7B-Instruct-v0.1.json b/models/mistralai__Mixtral-8x7B-Instruct-v0.1.json index 5558e40..8f0f7c4 100644 --- a/models/mistralai__Mixtral-8x7B-Instruct-v0.1.json +++ b/models/mistralai__Mixtral-8x7B-Instruct-v0.1.json @@ -69,6 +69,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": false, + "24gb": false, + "32gb": false, + "48gb": true, + "64gb": true, + "128gb": true + }, "mac_studio": { "32gb": false, "64gb": true, @@ -78,6 +86,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/models/unsloth__DeepSeek-V4-Flash-0731-GGUF.json b/models/unsloth__DeepSeek-V4-Flash-0731-GGUF.json index ebaef0f..aeba968 100644 --- a/models/unsloth__DeepSeek-V4-Flash-0731-GGUF.json +++ b/models/unsloth__DeepSeek-V4-Flash-0731-GGUF.json @@ -81,6 +81,14 @@ "96gb": true, "128gb": true }, + "mac_mini": { + "16gb": false, + "24gb": false, + "32gb": false, + "48gb": false, + "64gb": false, + "128gb": true + }, "mac_studio": { "32gb": false, "64gb": false, @@ -90,6 +98,11 @@ "256gb": true, "512gb": true }, + "mac_pro": { + "192gb": true, + "384gb": true, + "512gb": true + }, "dgx": { "640gb": true, "1128gb": true, diff --git a/openspec/changes/expand-apple-hardware/.openspec.yaml b/openspec/changes/expand-apple-hardware/.openspec.yaml new file mode 100644 index 0000000..d7bc011 --- /dev/null +++ b/openspec/changes/expand-apple-hardware/.openspec.yaml @@ -0,0 +1,2 @@ +schema: spec-driven +created: 2026-08-10 diff --git a/openspec/changes/expand-apple-hardware/README.md b/openspec/changes/expand-apple-hardware/README.md new file mode 100644 index 0000000..f9c94b0 --- /dev/null +++ b/openspec/changes/expand-apple-hardware/README.md @@ -0,0 +1,3 @@ +# expand-apple-hardware + +Mac Mini + Mac Pro hardware sections (unified memory) + agentic-coding capability flag (coding/reasoning + 32K context) diff --git a/openspec/changes/expand-apple-hardware/proposal.md b/openspec/changes/expand-apple-hardware/proposal.md new file mode 100644 index 0000000..8614c4a --- /dev/null +++ b/openspec/changes/expand-apple-hardware/proposal.md @@ -0,0 +1,24 @@ +# Mac Mini + Mac Pro sections + agentic-coding capability + +## Why + +The hardware list covered NVIDIA/AMD/Arc/Snapdragon/MacBook/Mac Studio/DGX/phones — +but the popular Apple desktops were missing: **Mac Mini** (M4 16 → M4 Max 128 GB +unified) and **Mac Pro** (M3 Ultra 192/384/512 GB). Users buying an agentic-coding +machine want to see these machines AND know whether a model can actually drive +agentic coding (coding/reasoning capability + long context). + +Verified 2026-08-10 (Apple spec pages + industry): Mac Mini M4 16 GB, M4 Pro +24/48 GB, M4 Max 32/64/128 GB; Mac Pro M3 Ultra 192/384/512 GB — both unified +memory (usable = tier − 3.5 GB, like all Macs). + +## Requirements + +- ADDED: `mac_mini` section (16/24/32/48/64/128) + `mac_pro` section + (192/384/512), unified-memory math, Metal backend, in estimator/schema/frontend + (hardware filter + wizard). +- ADDED: `agentic_capable(best_for, context_window)` — coding/reasoning/agentic + model AND ≥32K context; surfaced as a wizard checkbox filter + an "Agentic + coding ready" badge in the details pane and per wizard pick. +- Tests both sides: estimator tier pins + agentic unit tests; jsdom 11-section box + counts + Mac Mini/Pro wizard parity + agentic filter/badge. diff --git a/openspec/changes/expand-apple-hardware/specs/mac-mini-pro-agentic/spec.md b/openspec/changes/expand-apple-hardware/specs/mac-mini-pro-agentic/spec.md new file mode 100644 index 0000000..cbc35b1 --- /dev/null +++ b/openspec/changes/expand-apple-hardware/specs/mac-mini-pro-agentic/spec.md @@ -0,0 +1,34 @@ +## ADDED Requirements + +### Requirement: Mac Mini + Mac Pro sections +`MAC_MINI_TIERS` MUST be `[16, 24, 32, 48, 64, 128]` and `MAC_PRO_TIERS` MUST be +`[192, 384, 512]`; both MUST use unified-memory math (usable = tier − 3.5). The +schema MUST carry `mac_mini` and `mac_pro` flag sections; the frontend MUST list +both machines in the hardware filter and the wizard. + +#### Scenario: Mac Mini buyer +- **WHEN** a user picks Mac Mini 64 GB +- **THEN** usable memory is 60.5 GB and fit uses it (70B-class est 47.3 fits; 48 GB + does not) +- **AND** the section appears in the details pane with Metal backend label + +### Requirement: Agentic-coding capability +`agentic_capable(best_for, context_window)` MUST return True only for +coding/reasoning/agentic models with ≥ `AGENTIC_CTX_MIN` (32768) context. The +wizard MUST offer an "agentic coding" checkbox filter; the details pane MUST show +an "Agentic coding ready" badge for capable models; wizard picks MUST annotate +capable models. + +#### Scenario: agentic filter +- **WHEN** the agentic checkbox is checked +- **THEN** every pick is a coding/reasoning/agentic model with 32K+ context + +### Requirement: Hermetic tests +- Python: tier pins (Mac Mini 64 fits 70B, 48 does not; Mac Pro 192+ fits), + agentic_capable unit cases, schema sections. +- JS: 11 hardware sections with box counts (mac_mini 6, mac_pro 3); Mac Mini/Pro + wizard parity; agentic filter + badge. + +#### Scenario: gate runs +- **WHEN** CI runs pytest, jsdom and Playwright +- **THEN** all new assertions pass against the committed data diff --git a/openspec/changes/expand-apple-hardware/tasks.md b/openspec/changes/expand-apple-hardware/tasks.md new file mode 100644 index 0000000..6ad5d99 --- /dev/null +++ b/openspec/changes/expand-apple-hardware/tasks.md @@ -0,0 +1,7 @@ +## 1. Sections + agentic flag — PR-P + +- [ ] 1.1 `estimator.py`: MAC_MINI_TIERS/MAC_PRO_TIERS + flags + `agentic_capable()` + AGENTIC_CTX_MIN +- [ ] 1.2 `schemas/model.schema.json`: mac_mini + mac_pro sections; samples regenerate valid +- [ ] 1.3 Frontend: TIERS/SECTIONS/MAC_CATS, hw filter + wizard options, agentic checkbox + filter + details badge + pick annotation +- [ ] 1.4 Tests: pytest pins + agentic unit + schema; jsdom 11-section counts + Mac Mini/Pro parity + agentic filter/badge +- [ ] 1.5 README hardware list + openspec validate; gate green diff --git a/schemas/model.schema.json b/schemas/model.schema.json index 07c1d69..56d9c95 100644 --- a/schemas/model.schema.json +++ b/schemas/model.schema.json @@ -367,6 +367,58 @@ "32gb", "64gb" ] + }, + "mac_mini": { + "type": "object", + "additionalProperties": false, + "properties": { + "16gb": { + "type": "boolean" + }, + "24gb": { + "type": "boolean" + }, + "32gb": { + "type": "boolean" + }, + "48gb": { + "type": "boolean" + }, + "64gb": { + "type": "boolean" + }, + "128gb": { + "type": "boolean" + } + }, + "required": [ + "16gb", + "24gb", + "32gb", + "48gb", + "64gb", + "128gb" + ] + }, + "mac_pro": { + "type": "object", + "additionalProperties": false, + "properties": { + "192gb": { + "type": "boolean" + }, + "384gb": { + "type": "boolean" + }, + "512gb": { + "type": "boolean" + } + }, + "required": [ + "192gb", + "384gb", + "512gb" + ] } } }, diff --git a/src/whatllm/artifacts.py b/src/whatllm/artifacts.py index 3b422ae..49fb22a 100644 --- a/src/whatllm/artifacts.py +++ b/src/whatllm/artifacts.py @@ -20,6 +20,7 @@ "id", "name", "author", "parameters_b", "architecture", "pipeline_tag", "trending_score", "downloads", "est_vram_gb", "model_type", "commercial_ok", "license", "best_for", "servers", + "context_window", ] diff --git a/src/whatllm/estimator.py b/src/whatllm/estimator.py index 25fbec5..a6db105 100644 --- a/src/whatllm/estimator.py +++ b/src/whatllm/estimator.py @@ -41,6 +41,10 @@ # treated like macOS unified memory (usable = tier - MAC_SYSTEM_GB) MACBOOK_TIERS = [16, 24, 32, 48, 64, 96, 128] # M5 Max tops out at 128 GB (verified, Apple) MAC_STUDIO_TIERS = [32, 64, 96, 128, 192, 256, 512] # M4 Max 64, M3 Ultra up to 512 GB +MAC_MINI_TIERS = [16, 24, 32, 48, 64, 128] +# Mac Mini: M4 16, M4 Pro 24/48, M4 Max 32/64/128 GB (verified Apple spec pages) +MAC_PRO_TIERS = [192, 384, 512] +# Mac Pro: M3 Ultra configs 192/384/512 GB unified (verified Apple/industry) DGX_TIERS = [640, 1128, 1440] # DGX A100/H100 640, H200 1128, B200 1440 GB total ANDROID_TIERS = [8, 12, 16, 24] # e.g. Galaxy S25 Ultra 12/16 GB IPHONE_TIERS = [8, 12] # iPhone 17 Pro 12 GB @@ -137,6 +141,20 @@ def _phone_flags(params_b: float, est: float, min_est: float, return {"boxes": boxes, "note": note, "practical": practical} +AGENTIC_CTX_MIN = 32768 # agentic coding needs long context (heuristic, documented) + + +def agentic_capable(best_for: list[str] | None, context_window: int | None) -> bool: + """A model can drive agentic coding when it is a coding/reasoning/agentic + model AND offers long context (>= AGENTIC_CTX_MIN).""" + if not best_for: + return False + tags = {t.lower() for t in best_for} + if not (tags & {"agentic", "coding", "reasoning", "code"}): + return False + return isinstance(context_window, int) and context_window >= AGENTIC_CTX_MIN + + def hardware_flags(params_b: float, quants: list[dict]) -> dict: """Hardware compatibility anchored on the recommended quant (quants[0]).""" if not quants: @@ -149,15 +167,18 @@ def hardware_flags(params_b: float, quants: list[dict]) -> dict: intel_arc = _tier_flags(est, INTEL_ARC_TIERS) snapdragon = _mac_flags(est, SNAPDRAGON_TIERS) # unified memory like macOS macbook = _mac_flags(est, MACBOOK_TIERS) + mac_mini = _mac_flags(est, MAC_MINI_TIERS) mac_studio = _mac_flags(est, MAC_STUDIO_TIERS) + mac_pro = _mac_flags(est, MAC_PRO_TIERS) dgx = _tier_flags(est, DGX_TIERS) android = _phone_flags(params_b, est, min_est, ANDROID_TIERS, "Android") iphone = _phone_flags(params_b, est, min_est, IPHONE_TIERS, "iPhone") consumer_any = ( any(nvidia.values()) or any(amd.values()) or any(intel_arc.values()) - or any(snapdragon.values()) or any(macbook.values()) - or any(mac_studio.values()) or any(android["boxes"].values()) + or any(snapdragon.values()) or any(macbook.values()) or any(mac_mini.values()) + or any(mac_studio.values()) or any(mac_pro.values()) + or any(android["boxes"].values()) or any(iphone["boxes"].values()) ) if not consumer_any: @@ -171,7 +192,9 @@ def hardware_flags(params_b: float, quants: list[dict]) -> dict: "intel_arc": intel_arc, "snapdragon": snapdragon, "macbook": macbook, + "mac_mini": mac_mini, "mac_studio": mac_studio, + "mac_pro": mac_pro, "dgx": dgx, "android": {**android["boxes"], "note": android["note"]}, "iphone": {**iphone["boxes"], "note": iphone["note"]}, diff --git a/tests/frontend.test.mjs b/tests/frontend.test.mjs index 4145af3..981567e 100644 --- a/tests/frontend.test.mjs +++ b/tests/frontend.test.mjs @@ -94,13 +94,15 @@ test('selecting a model renders 7 hardware sections with correct box counts', as click(window, '.card'); await waitFor(() => window.document.querySelector('#d-detail').style.display !== 'none', 6000); const h3s = [...window.document.querySelectorAll('#d-hw-sections h3')].map((h) => h.textContent); - assert.equal(h3s.length, 9); + assert.equal(h3s.length, 11); assert.equal(window.document.querySelectorAll('#hw-nvidia .box').length, 8); assert.equal(window.document.querySelectorAll('#hw-amd .box').length, 8); assert.equal(window.document.querySelectorAll('#hw-intel_arc .box').length, 4); assert.equal(window.document.querySelectorAll('#hw-snapdragon .box').length, 3); assert.equal(window.document.querySelectorAll('#hw-macbook .box').length, 7); + assert.equal(window.document.querySelectorAll('#hw-mac_mini .box').length, 6); assert.equal(window.document.querySelectorAll('#hw-mac_studio .box').length, 7); + assert.equal(window.document.querySelectorAll('#hw-mac_pro .box').length, 3); assert.equal(window.document.querySelectorAll('#hw-dgx .box').length, 3); assert.equal(window.document.querySelectorAll('#hw-android .box').length, 4); assert.equal(window.document.querySelectorAll('#hw-iphone .box').length, 2); @@ -447,3 +449,62 @@ test('Bootstrap theme: vendored stylesheet linked + wizard shows best-server sum assert.ok(styles.includes('.pcols') && styles.includes('.cmptable'), 'Quartz overrides in style block'); dom.window.close(); }); + +test('Mac Mini + Mac Pro sections appear in hardware filter and wizard (issue: more Macs)', async () => { + const dom = await boot(); + const { window } = dom; + await waitFor(() => cardCount(window) > 0); + const hwOpts = [...window.document.querySelectorAll('#hwcat option')].map(o => o.value); + assert.ok(hwOpts.includes('mac_mini') && hwOpts.includes('mac_pro'), 'hw filter has Mac Mini/Pro'); + const wOpts = [...window.document.querySelectorAll('#w-hwcat option')].map(o => o.value); + assert.ok(wOpts.includes('mac_mini') && wOpts.includes('mac_pro'), 'wizard has Mac Mini/Pro'); + change(window, '#w-hwcat', 'mac_mini'); + await waitFor(() => !window.document.querySelector('#w-hwtier').disabled); + change(window, '#w-hwtier', '64'); + click(window, '#w-go'); + await waitFor(() => window.document.querySelectorAll('#w-results .wrow').length > 0); + [...window.document.querySelectorAll('#w-results .wrow')].forEach(r => { + const name = r.querySelector('a').textContent; + const m = window.MODELS_INDEX.find(x => name.startsWith(x.name)); + assert.ok(m && m.est_vram_gb + 1.5 <= 64 - 3.5, name + ' fits Mac Mini 64GB (unified 60.5)'); + }); + dom.window.close(); +}); + +test('agentic-coding filter + badge (coding/reasoning model with 32K+ context)', async () => { + const dom = await boot(); + const { window } = dom; + await waitFor(() => cardCount(window) > 0); + change(window, '#w-hwcat', 'nvidia'); + await waitFor(() => !window.document.querySelector('#w-hwtier').disabled); + change(window, '#w-hwtier', '24'); + // without the filter: picks exist + click(window, '#w-go'); + await waitFor(() => window.document.querySelectorAll('#w-results .wrow').length > 0); + // with the agentic filter: every pick is a coding/reasoning model with 32K+ ctx + check(window, '#w-agentic', true); + click(window, '#w-go'); + await waitFor(() => window.document.querySelectorAll('#w-results .wrow').length > 0); + [...window.document.querySelectorAll('#w-results .wrow')].forEach(r => { + const name = r.querySelector('a').textContent; + const m = window.MODELS_INDEX.find(x => name.startsWith(x.name)); + const tags = (m.best_for || []).join(' ').toLowerCase(); + assert.ok(/(agentic|coding|reasoning|code)/.test(tags), name + ' is coding/reasoning'); + assert.ok((m.context_window || 0) >= 32768, name + ' has 32K+ context'); + }); + // details pane shows the agentic badge for a capable model + click(window, '.card'); + await waitFor(() => window.document.querySelector('#d-detail').style.display !== 'none', 6000); + const badge = window.document.querySelector('#d-agentic'); + const sel = window.MODELS_INDEX.find(x => x.id === window.document.querySelector('.card.sel')?.dataset?.id); + if (sel && agenticTest(sel)) { + assert.ok(badge && badge.style.display !== 'none' && badge.textContent.includes('Agentic coding ready'), + 'agentic badge visible: ' + (badge && badge.textContent.slice(0,60))); + } + dom.window.close(); +}); + +function agenticTest(m){ + const tags = (m.best_for || []).join(' ').toLowerCase(); + return /(agentic|coding|reasoning|code)/.test(tags) && (m.context_window || 0) >= 32768; +} diff --git a/tests/test_estimator.py b/tests/test_estimator.py index 6e3c775..a498847 100644 --- a/tests/test_estimator.py +++ b/tests/test_estimator.py @@ -233,3 +233,27 @@ def test_flags_snapdragon_unified_memory_math(): hw_big = estimator.hardware_flags(85.0, qs_big) assert hw_big["snapdragon"]["32gb"] is False # 57.8 > 28.5 assert hw_big["snapdragon"]["64gb"] is True # 57.8 <= 60.5 (64-3.5) + + +def test_flags_mac_mini_and_mac_pro_sections(): + """Mac Mini (M4 16..M4 Max 128) + Mac Pro (M3 Ultra 192/384/512), unified -3.5.""" + qs = estimator.synthesize_quants(72.71) # est 45.8 -> needs 47.3 + hw = estimator.hardware_flags(72.71, qs) + assert hw["mac_mini"]["48gb"] is False # 47.3 > 44.5 (48-3.5) + assert hw["mac_mini"]["64gb"] is True # 47.3 <= 60.5 + assert hw["mac_mini"]["128gb"] is True + assert hw["mac_pro"]["192gb"] is True + assert hw["mac_pro"]["512gb"] is True + # a small model fits the base Mac Mini + hw2 = estimator.hardware_flags(8.19, estimator.synthesize_quants(8.19)) + assert hw2["mac_mini"]["16gb"] is True + + +def test_agentic_capable(): + assert estimator.agentic_capable(["coding", "chat"], 32768) is True + assert estimator.agentic_capable(["reasoning"], 131072) is True + assert estimator.agentic_capable(["agentic"], 32768) is True + assert estimator.agentic_capable(["coding"], 8192) is False # context too small + assert estimator.agentic_capable(["chat", "rag"], 131072) is False # no coding tag + assert estimator.agentic_capable(None, 131072) is False + assert estimator.agentic_capable(["coding"], None) is False diff --git a/tests/test_schema.py b/tests/test_schema.py index ee1aa27..7d3d33a 100644 --- a/tests/test_schema.py +++ b/tests/test_schema.py @@ -37,6 +37,9 @@ "64gb": True, "96gb": True, "128gb": True}, "mac_studio": {"32gb": True, "64gb": True, "96gb": True, "128gb": True, "192gb": True, "256gb": True, "512gb": True}, + "mac_mini": {"16gb": True, "24gb": True, "32gb": True, "48gb": True, + "64gb": True, "128gb": True}, + "mac_pro": {"192gb": True, "384gb": True, "512gb": True}, "dgx": {"640gb": True, "1128gb": True, "1440gb": True}, "android": {"8gb": True, "12gb": True, "16gb": True, "24gb": True, "note": "Practical at the smallest quant (~3.6 GB est.) on flagship Android"}, @@ -177,3 +180,15 @@ def test_nvidia_32gb_tier_valid(): rec = json.loads(json.dumps(VALID)) rec["hardware"]["nvidia"]["32gb"] = True # RTX 5090 tier key assert validate(rec) == [] + + +def test_mac_mini_rejects_unknown_tier(): + rec = json.loads(json.dumps(VALID)) + rec["hardware"]["mac_mini"]["96gb"] = True # no 96 GB Mac Mini config + assert validate(rec) + + +def test_mac_pro_requires_all_tiers(): + rec = json.loads(json.dumps(VALID)) + rec["hardware"]["mac_pro"] = {"192gb": True} # missing 384gb/512gb + assert validate(rec)