Leaderboard
Bradley–Terry scores from 8 blind votes, with 95% confidence intervals. Models whose intervals overlap share a rank. How it works
| Rank | Model | Score | 95% CI | Votes | Win rate |
|---|---|---|---|---|---|
| 1 | GPT-6.1 Sol OpenAI | 1684 | 5 | 100% | |
| 1 | Qwen 3.8 Max Alibaba | 1592 | 1 | 100% | |
| 1 | GPT-6 Luna OpenAI | 1458 | 2 | 50% | |
| 2 | Claude Haiku 5.5 Anthropic | 1454 | 3 | 0% | |
| 2 | Claude Sonnet 5.5 Anthropic | 1417 | 4 | 25% | |
| 2 | Grok 4.7 xAI | 1408 | 1 | 0% |
Not rated yet: Claude Opus 5.5, Gemini 4 Argon, Gemini 3.8 Flash, DeepSeek V4.1 Flash, Kimi K3.1 Pro, GLM 5.3, MiniMax M3.1 Flash
updated 2026-10-25 15:59 UTC