RewardBench 2 Ties: leaderboard
RewardBench 2 ties slice checks that reward models rank every valid answer above incorrect ones when a prompt has several correct responses.
Metric: Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 197 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 85.42 | #492 |
| 2 | Claude Opus 4 (20250514) | 83.75 | #276 |
| 3 | Gemini 2.5 Flash (Preview 04-17) | 83.41 | #449 |
| 4 | Gemini 2.5 Pro | 81.1 | #262 |
| 5 | Gemini 2.5 Flash | 80.9 | #452 |
| 6 | Claude Sonnet 4 (20250514) | 79.39 | #386 |
| 7 | GPT-4o (2024-08-06) | 78.19 | #731 |
| 8 | GPT-4.1 Mini | 74 | #775 |
| 9 | Gemini 2.5 Pro (Preview 05-06) | 69.73 | #271 |
| 10 | GPT-4o Mini (2024-07-18) | 69.62 | #1512 |
| 11 | Claude 3.5 Sonnet (20240620) | 67.4 | #684 |
| 12 | Claude 3.7 Sonnet (20250219) | 67.23 | #399 |
| 13 | Claude 3 Opus (20240229) | 56.01 | #1116 |
| 14 | GPT-4.1 Nano | 50.15 | #1992 |
| 15 | Gemini 1.5 Flash-8B | 24.21 | #2652 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rewardbench-2-ties · How It Works · Data refreshed daily, snapshot 2026-11-12.