Skip to content

MODEL-RESEARCH.md

Latest commit

 

History

History
271 lines (216 loc) · 8.17 KB

MODEL-RESEARCH.md

File metadata and controls

271 lines (216 loc) · 8.17 KB

Thinking Agents — Model Research

Peripheral tier: cheap background thinking ticks (~500 token prompt, ~200 token response) Research date: 2026-02-15

OpenClaw Provider Support Summary

Model Built-in Provider Custom Config Needed
Kimi (Moonshot) ❌ Not built-in ✅ Yes — models.providers with OpenAI-compatible baseUrl
GLM (Zhipu/Z.AI) ✅ Built-in (zai) No — just set ZAI_API_KEY
Qwen (Alibaba) ✅ Built-in (qwen-portal plugin) No — OAuth device flow or DashScope API key
DeepSeek ✅ Via HuggingFace or OpenRouter Custom provider also works (OpenAI-compatible)
Mistral ✅ Built-in (mistral) No — just set MISTRAL_API_KEY
Ollama (local) ✅ Built-in (ollama) No — auto-detected at localhost:11434

1. Kimi (Moonshot AI)

  • Sign up: https://platform.moonshot.ai — email registration, $5 bonus on first $5 recharge
  • API Endpoint: https://api.moonshot.ai/v1 (OpenAI-compatible)
  • Auth: Bearer token (MOONSHOT_API_KEY)
  • OpenAI-compatible: ✅ Yes — chat completions format
  • Models: kimi-k2.5, kimi-k2-0905-preview, kimi-k2-turbo-preview, kimi-k2-thinking
  • Context: up to 131K–256K tokens

Pricing (per 1M tokens)

Model Input Output
Kimi K2 $0.50–$0.60 $2.40–$2.50
Kimi K2 (cache hit) $0.15 $2.50

Cost per thinking tick

  • Input: 500 tokens × $0.60/1M = $0.0003
  • Output: 200 tokens × $2.50/1M = $0.0005
  • Total: ~$0.0008/tick ($0.80 per 1000 ticks)

OpenClaw Config

{
  models: {
    mode: "merge",
    providers: {
      moonshot: {
        baseUrl: "https://api.moonshot.ai/v1",
        apiKey: "${MOONSHOT_API_KEY}",
        api: "openai-completions",
        models: [{ id: "kimi-k2.5", name: "Kimi K2.5" }],
      },
    },
  },
}

2. GLM (Zhipu AI / Z.AI)

  • Sign up: https://open.bigmodel.cn or https://z.ai
  • API Endpoint: https://open.bigmodel.cn/api/paas/v4 (or Z.AI endpoint)
  • Auth: Bearer token (ZAI_API_KEY)
  • OpenAI-compatible: ✅ Yes (also supports Anthropic format)
  • Context: up to 200K tokens (GLM-4.7)

Pricing (per 1M tokens)

Model Input Output
GLM-4.7 $0.60 $2.20
GLM-4.7-FlashX $0.07 $0.40
GLM-4.5-Air $0.20 $1.10
GLM-4.7-Flash FREE FREE
GLM-4.5-Flash FREE FREE

Cost per thinking tick

  • GLM-4.7-Flash: $0.00/tick (FREE!) ← Best for thinking ticks
  • GLM-4.7-FlashX: ~$0.0001/tick
  • GLM-4.7: ~$0.0007/tick

OpenClaw Config (built-in)

{
  agents: { defaults: { model: { primary: "zai/glm-4.7-flash" } } },
}
// Just set env: ZAI_API_KEY

⭐ RECOMMENDED for free thinking ticks — GLM-4.7-Flash and GLM-4.5-Flash are completely free.

3. Qwen (Alibaba Cloud / DashScope)

  • Sign up: https://dashscope.console.aliyun.com or https://qwen.ai/apiplatform
  • API Endpoint: https://dashscope.aliyuncs.com/compatible-mode/v1 (OpenAI-compatible)
  • Auth: Bearer token (DASHSCOPE_API_KEY)
  • OpenAI-compatible: ✅ Yes — explicit OpenAI Chat Completion compatibility
  • Also available via: OpenClaw qwen-portal plugin (OAuth, free tier)

Pricing (per 1M tokens)

Model Input Output
Qwen-Max $0.38 ~$1.20
Qwen-Plus ~$0.15 ~$0.60
Qwen-Turbo ~$0.05 ~$0.20
Qwen3-32B (via Groq) $0.29 ~$0.39

Cost per thinking tick

  • Qwen-Turbo: ~$0.0001/tick
  • Qwen-Plus: ~$0.0002/tick
  • Qwen-Max: ~$0.0004/tick

OpenClaw Config

Option A — Built-in plugin (free OAuth tier):

openclaw plugins enable qwen-portal-auth
openclaw models auth login --provider qwen-portal --set-default
# Model: qwen-portal/coder-model

Option B — DashScope API key (custom provider):

{
  models: {
    mode: "merge",
    providers: {
      qwen: {
        baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
        apiKey: "${DASHSCOPE_API_KEY}",
        api: "openai-completions",
        models: [{ id: "qwen-turbo", name: "Qwen Turbo" }],
      },
    },
  },
}

4. DeepSeek (V3 / R1)

  • Sign up: https://platform.deepseek.com
  • API Endpoint: https://api.deepseek.com/v1 (OpenAI-compatible)
  • Auth: Bearer token (DEEPSEEK_API_KEY)
  • OpenAI-compatible: ✅ Yes
  • Context: 64K tokens

Pricing (per 1M tokens)

Model Input (cache miss) Input (cache hit) Output
deepseek-chat (V3) $0.27 $0.07 $1.10
deepseek-reasoner (R1) $0.55 $0.14 $2.19

Cost per thinking tick

  • DeepSeek V3: 500×$0.27/1M + 200×$1.10/1M = ~$0.0004/tick
  • DeepSeek V3 (cached): ~$0.0003/tick
  • DeepSeek R1: ~$0.0007/tick (plus CoT tokens add cost)

OpenClaw Config

{
  models: {
    mode: "merge",
    providers: {
      deepseek: {
        baseUrl: "https://api.deepseek.com/v1",
        apiKey: "${DEEPSEEK_API_KEY}",
        api: "openai-completions",
        models: [
          { id: "deepseek-chat", name: "DeepSeek V3" },
          { id: "deepseek-reasoner", name: "DeepSeek R1" },
        ],
      },
    },
  },
}

Also available via: huggingface/deepseek-ai/DeepSeek-R1 (built-in) or OpenRouter.

5. Mistral (Small / Large)

  • Sign up: https://console.mistral.ai
  • API Endpoint: https://api.mistral.ai/v1 (OpenAI-compatible)
  • Auth: Bearer token (MISTRAL_API_KEY)
  • OpenAI-compatible: ✅ Yes
  • OpenClaw built-in: ✅ Yes (mistral provider)

Pricing (per 1M tokens)

Model Input Output
Mistral Small $1.00 $3.00
Mistral Large $2.00 $6.00

Cost per thinking tick

  • Mistral Small: 500×$1.00/1M + 200×$3.00/1M = ~$0.0011/tick
  • Mistral Large: ~$0.0022/tick

OpenClaw Config (built-in)

export MISTRAL_API_KEY="your-key"
# Model: mistral/mistral-small-latest or mistral/mistral-large-latest

6. Ollama (Local — Llama 3 8B)

  • Install: https://ollama.ai → curl -fsSL https://ollama.ai/install.sh | sh
  • API Endpoint: http://127.0.0.1:11434/v1 (OpenAI-compatible)
  • Auth: None
  • OpenAI-compatible: ✅ Yes
  • Cost: $0.00 (local compute only — electricity + hardware)
  • RAM needed: ~5GB for Llama 3 8B (quantized)

Setup

ollama pull llama3:8b
# Or for newer: ollama pull llama3.3

Cost per thinking tick

  • $0.00/tick ← Zero marginal cost

OpenClaw Config (built-in, auto-detected)

{
  agents: { defaults: { model: { primary: "ollama/llama3:8b" } } },
}

Hardware Note

Our server (2-core Xeon, 8GB RAM) can run Llama 3 8B quantized (Q4_K_M) but expect ~2-5 tokens/sec. Adequate for background thinking ticks, not for interactive use.

Comparison Matrix

Provider Cost/Tick OpenAI Compat Built-in OC Quality Latency
GLM-4.7-Flash $0.000 ✅ ✅ Medium Fast
Ollama Llama3 8B $0.000 ✅ ✅ Low-Med Slow (local)
Qwen-Turbo $0.0001 ✅ ✅ plugin Medium Fast
DeepSeek V3 $0.0004 ✅ Via proxy High Fast
Kimi K2 $0.0008 ✅ Custom High Medium
Mistral Small $0.0011 ✅ ✅ Medium Fast
GLM-4.7 $0.0007 ✅ ✅ High Fast
DeepSeek R1 $0.0007+ ✅ Via proxy Very High Slow (CoT)
Mistral Large $0.0022 ✅ ✅ High Fast

Recommendation

Tier 1 — Free thinking ticks:

  1. GLM-4.7-Flash / GLM-4.5-Flash (Z.AI) — Free, cloud-hosted, built-in OpenClaw support. Best option.
  2. Ollama Llama 3 8B — Free, local, but slow on our hardware.

Tier 2 — Near-free (<$1/1000 ticks): 3. Qwen-Turbo — $0.10/1000 ticks, OpenClaw plugin available 4. DeepSeek V3 — $0.40/1000 ticks, excellent quality/price ratio

Tier 3 — Budget ($1-2/1000 ticks): 5. Kimi K2 — $0.80/1000 ticks, strong reasoning 6. Mistral Small — $1.10/1000 ticks, built-in support

All providers use OpenAI-compatible API format. Integration with OpenClaw is straightforward for all of them.