Free LLM APIs · state
The state of free LLM APIs
Free chat models on OpenRouter, the largest catalogue of free LLM endpoints, scored from published benchmarks and LM Arena ratings, with who serves each one. Computed from the open ranking, last updated 22 Oct 2026.
What the data says today
- 8 of 17 ranked free models are served only by providers that may train on your prompts; 6 are served by providers that don't. Provider policies as published by OpenRouter, read 2026-10-10.
- The best-scoring free model from a provider that doesn't train is Google: Gemma 4 31B (#4, score 57.1).
- Only 12 of 17 have published benchmark indices or LM Arena ratings; the rest are scored by an estimate, capped low, until they're measured.
Top 10 free models
| # | Model | Score | Served by | Prompts |
|---|---|---|---|---|
| 1 | Thinking Machines: Inkling thinkingmachines/inkling:free | 67.7 | Thinking Machines | May train |
| 2 | Thinking Machines: Inkling Small thinkingmachines/inkling-small:free | 64.6 | Thinking Machines | May train |
| 3 | MiniMax: MiniMax M3.1 Flash minimax/minimax-m3.1-flash:free | 57.5 | Minimax | Unknown |
| 4 | Google: Gemma 4 31B google/gemma-4-31b-it:free | 57.1 | Google AI Studio | No training |
| 5 | Google: Gemma 4 26B A4B google/gemma-4-26b-a4b-it:free | 55 | Google AI Studio | No training |
| 6 | NVIDIA: Nemotron 3 Super nvidia/nemotron-3-super-120b-a12b:free | 45.5 | Nvidia | May train |
| 7 | NVIDIA: Nemotron 3 Ultra nvidia/nemotron-3-ultra-550b-a55b:free | 44.4 | Nvidia | May train |
| 8 | NVIDIA: Nemotron 3.5 Super VL nvidia/nemotron-3.5-super-vl:free | 43.3 | Nvidia | May train |
| 9 | Poolside: Laguna S 2.1 poolside/laguna-s-2.1:free | 33.8 | Poolside | No training |
| 10 | Poolside: Laguna XS 2.1 poolside/laguna-xs-2.1:free | 26 | Poolside | No training |
Score 0-100 from Artificial Analysis indices and LM Arena ratings; method in scoring/. Our live leaderboard also weighs how each model is answering right now.
Free models listed over time
Free model ids on OpenRouter, including the occasional non-chat model. One snapshot a day in the repo's data/history.
All of this is open: the scoring code, the inputs behind every score and a snapshot per day. If it's useful, a star helps other people find it.
Star on GitHubLimits and fine print per provider: free LLM APIs compared.