RewardBench: leaderboard
Reward model benchmark evaluating preference models across chat, hard chat, safety, reasoning, and prior preference-evaluation sets.
Metric: Score (%). Source: github.com. Source: huggingface.co. Saturation forecast: Estimated already saturated. 178 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | ArmoRM-Llama3-8B-v0.1 | 88.6 | #1394 |
| 2 | Gemini 1.5 Pro (0514) | 88.2 | #512 |
| 3 | Gemini 1.5 Pro (Sept) | 86.78 | #448 |
| 4 | GPT-4o (2024-08-06) | 86.73 | #336 |
| 5 | GPT-4 Preview (0125) | 84.34 | #486 |
| 6 | Claude 3.5 Sonnet (20240620) | 84.17 | #299 |
| 7 | Llama 3.1 405B Instruct | 84.12 | #457 |
| 8 | Llama 3.1 70B Instruct | 84.05 | #561 |
| 9 | GPT-4 Turbo | 83.95 | #404 |
| 10 | GPT-4o (2024-05-13) | 83.27 | #325 |
| 11 | Gemma 2 27B (IT) | 80.9 | #645 |
| 12 | Gemini 1.5 Flash (001) | 80.54 | #577 |
| 13 | Claude 3 Opus (20240229) | 80.08 | #412 |
| 14 | GPT-4o Mini (2024-07-18) | 80.07 | #558 |
| 15 | Hermes 3 - Llama-3.1 70B | 78.47 | #927 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=rewardbench · How It Works · Data refreshed daily, snapshot 2026-10-20.