LLM benchmarks
Every model's score on every benchmark we track, each linked to where it was published. Independent runs are listed apart from the numbers labs report for their own models. Many of them feed the Respan Index.
Every model's score on every benchmark we track, each linked to where it was published. Independent runs are listed apart from the numbers labs report for their own models. Many of them feed the Respan Index.
Every model's score on every benchmark we track, each linked to where it was published. Independent runs are listed apart from the numbers labs report for their own models. Many of them feed the Respan Index.
Every model's score on every benchmark we track, each linked to where it was published. Independent runs are listed apart from the numbers labs report for their own models. Many of them feed the Respan Index.
Models ranked on one 0 to 100 score from public benchmarks and lab results.
161 ranked modelsVersion 1.1How it is scored
| Model | Respan Index | |
|---|---|---|
| 1. | Claude Opus 5.5by Anthropic | 73.7 |
| 2. | Claude Sonnet 5.5by Anthropic | 72.9 |
| 3. | Claude Fable 5.1by Anthropic | 72.1 |
| 3. | GPT-6 Astraby OpenAI | 72.1 |
| 5. | Kimi K3.1 Proby Moonshot AI | 71.2 |
| 6. | Muse Wonder Fullby Meta | 71.0 |
| 7. | GPT-6.1 Solby OpenAI | 69.8 |
| 8. | Gemini 4 Argonby Google | 69.1 |
| 9. | GPT-6.1 Terraby OpenAI | 68.9 |
| 10. | Claude Opus 5by Anthropic | 67.1 |
| 11. | GPT-6 Solby OpenAI | 64.7 |
| 12. | Claude Fable 5by Anthropic | 63.4 |
| 13. | GPT-5.6 Solby OpenAI | 61.6 |
| 14. | Muse Wonderby Meta | 61.1 |
| 15. | DeepSeek V4.1 Flashby DeepSeek | 57.4 |
| 16. | Claude Haiku 5.5by Anthropic | 57.0 |
| 17. | MiMo-V2.6-Proby Xiaomi MiMo | 56.3 |
| 18. | Claude Opus 4.8by Anthropic | 55.8 |
| 19. | Step 5 Previewby StepFun | 55.7 |
| 20. | Muse Spark 1.3by Meta | 55.6 |
| 21. | GPT-5.6 Terraby OpenAI | 55.2 |
| 22. | Gemini 3.8 Flashby Google | 54.6 |
| 23. | GPT-5.5by OpenAI | 54.1 |
| 24. | Grok 4.6by xAI | 53.6 |
| 25. | Kimi K3by Moonshot AI | 53.1 |
* Estimated, no published scores in that category.
A human-preference leaderboard: people compare two anonymous models' answers blind and vote, producing an Elo rating from millions of pairwise votes.
Arena163 modelsRun by huggingface.co/lmarena-ai, Oct 28, 2026