Skip to content

Add free-best-router to Models & Providers - #125

Open
MajidAsghariTabrizi wants to merge 1 commit into
Anil-matcha:mainfrom
MajidAsghariTabrizi:add-free-best-router
Open

MajidAsghariTabrizi wants to merge 1 commit into
Anil-matcha:mainfrom
MajidAsghariTabrizi:add-free-best-router

Conversation

@MajidAsghariTabrizi

@MajidAsghariTabrizi MajidAsghariTabrizi commented Sep 3, 2026

Copy link
Copy Markdown

Adds free-best-router to the Models & Providers section.

It is an OpenAI-compat routing layer across 7 free LLM providers (OpenRouter, OpenCode/Zen, Groq, Cerebras, Mistral, DeepSeek, and local Ollama / LM Studio / llama.cpp) with:

  • Capability-aware ranking (7 weighted components: capability fit, agentic quality, reliability, coding quality, context fit, tool quality, latency)
  • Wilson lower bound on success rate plus time-decay penalties
  • Per-failure-type cooldowns (60s base for 429, 30-min base for 404 with 1h cap, exponential backoff)
  • Bounded failover: up to 4 model attempts, 45s per attempt, with Retry-After and next_eligible_model headers when the pool is exhausted
  • 5% smart exploration restricted to healthy, capability-close runners-up

DSH integration is a 6-line YAML config block (no native plugin needed). Full write-up in the DSH "Show Your Plugins!" discussion #5508.

Entry matches the existing one-line format used in this section.

Disclosure: this is my own open-source project. I rewrote the entry to use capability-based wording only (no hard-coded metric counts), per the maintainer's earlier feedback on similar entries.

Adds [free-best-router](https://github.com/MajidAsghariTabrizi/free-best-router) to the Models & Providers section. Capability-based wording, no hard-coded test counts. MIT, zero telemetry.

Disclosure: this is my own open-source project. It fits this list as an OpenAI-compat routing layer for the free tiers of OpenRouter, OpenCode/Zen, Groq, Cerebras, Mistral, DeepSeek, and local Ollama / LM Studio / llama.cpp. DSH integration uses the existing OpenAI-compatible provider config (no native plugin needed); full write-up at deepseek-ai/deepseek-harness#5508.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant