FormalRewardBench — leaderboard
Metric: Pairwise Accuracy (self-reported). Source: benchmarklist.com. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 59.8 |
| 2 | Claude Sonnet 4.5 | 45.7 |
| 3 | GPT-5.2 | 39.7 |
| 4 | Qwen 2.5 Coder 32B | 37.6 |
| 5 | Claude Sonnet 4 | 32.4 |
| 6 | GPT-4.1 | 32.1 |
| 7 | GPT-5.1 | 30.2 |
| 8 | Gemini 2.5 Flash | 25.2 |
| 9 | GPT-4o | 23.1 |
| 10 | Qwen2.5-Math-7B | 0 |
Interactive version: theaggregate.ai/benchmark?slug=formalrewardbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.