RewardBench Math — leaderboard
Metric: Score (%). Source: huggingface.co. 197 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 89.8 |
| 2 | Gemini 2.5 Flash | 85.2 |
| 3 | Gemini 2.5 Flash (Preview 04-17) | 81.15 |
| 4 | Claude 3.7 Sonnet (20250219) | 75 |
| 5 | Claude Opus 4 (20250514) | 74.91 |
| 6 | GPT-4.1 Mini | 72.13 |
| 7 | Claude Sonnet 4 (20250514) | 70.49 |
| 8 | Skywork-Reward-Gemma-2-27B-v0.2 | 67.21 |
| 9 | ArmoRM-Llama3-8B-v0.1 | 66.12 |
| 10 | GPT-4.1 | 65.21 |
| 11 | Llama-3.1-Tulu-3-8B-RM | 64.48 |
| 12 | GPT-4o (2024-08-06) | 62.3 |
| 13 | Claude 3.5 Sonnet (20240620) | 56.83 |
| 14 | Gemini 2.5 Pro (Preview 05-06) | 53.42 |
| 15 | GPT-4o Mini (2024-07-18) | 51.91 |
Interactive version: theaggregate.ai/benchmark?slug=rewardbench-math · How the rankings work · Data refreshed daily, snapshot 2026-07-22.