Arena-Hard v2 (GPT-4.1 Judge) — leaderboard
500 challenging real-world queries auto-evaluated by GPT-4.1 as judge, with style control. Alternative judging config for Arena-Hard v2.
Metric: Win Rate (%). Source: github.com. Status: saturation imminent. 28 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 (2025-04-16) | 87 |
| 2 | O4 Mini (2025-04-16) (High) | 81.7 |
| 3 | O4 Mini (2025-04-16) | 78 |
| 4 | O3 Mini (2025-01-31) (High) | 64.8 |
| 5 | O1 (2024-12-17) (High) | 58.7 |
| 6 | GPT-4.1 | 58.3 |
| 7 | O1 (2024-12-17) | 50.2 |
| 8 | O3 Mini (2025-01-31) | 50 |
| 9 | GPT-4.1 Mini | 48.6 |
| 10 | DeepSeek R1 | 48 |
| 11 | Qwen 3 235B A22B | 46.7 |
| 12 | Gemini 2.5 Flash | 45.1 |
| 13 | GPT-4.5 (Preview) | 43 |
| 14 | QwQ-32B | 36.1 |
| 15 | Qwen 3 32B | 35.8 |
Interactive version: theaggregate.ai/benchmark?slug=arena-hard-v2-gpt-4-1-judge · How the rankings work · Data refreshed daily, snapshot 2026-07-22.