RewardBench 2 Focus — leaderboard
Metric: Accuracy (%). Source: huggingface.co. 52 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek-V4-Flash-FP8 | 93.64 |
| 2 | DeepSeek R1 0528 | 93.59 |
| 3 | DeepSeek V4 Pro | 93.13 |
| 4 | DeepSeek V3.2 | 92.63 |
| 5 | Kimi K2.6 | 92.53 |
| 6 | DeepSeek V3.1 Terminus | 92.42 |
| 7 | MiMo-V2-Flash | 92.12 |
| 8 | Kimi K2.5 | 91.72 |
| 9 | NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 91.11 |
| 10 | GLM-4.6 FP8 | 90.91 |
| 11 | Qwen3-VL-235B-A22B-Thinking-FP8 | 90.91 |
| 12 | GLM-4.7 FP8 | 90.51 |
| 13 | Mistral-Large-3-675B-Instruct-2512 | 90.51 |
| 14 | GPT-OSS-120B | 90.35 |
| 15 | DeepSeek V3.2 Speciale | 90.35 |
Interactive version: theaggregate.ai/benchmark?slug=rewardbench-2-focus · How the rankings work · Data refreshed daily, snapshot 2026-07-22.