MultiChallenge - OpenAI o3-mini grader
Chat & Writing · 2025-08-07
Scale AI's MultiChallenge multi-turn conversations as scored by OpenAI with o3-mini as the judge in place of the benchmark's default GPT-4o grader, which OpenAI found frequently mis-scored responses. Not comparable with Scale's leaderboard scores; diagnostic. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5 | 69.6 |
| GPT-5 Mini | 62.3 |
| o3 | 60.4 |
| O4 Mini | 57.5 |
| GPT-5 Chat | 54.9 |
| GPT-5 Nano | 54.9 |
| o1 (2024-12-17) | 52.9 |
| o3-mini | 50.2 |
| GPT-4.5 | 50.1 |
| GPT-4.1 | 46.2 |
| GPT-4.1 Mini | 42.2 |
| GPT-4o (2024-11-20) | 39.9 |
| GPT-4.1 Nano | 31.1 |
| GPT-4o Mini | 25.6 |