Themis-CodeRewardBench: leaderboard
Metric: Preference accuracy (%; share of the benchmark's preference pairs in which the reward model scores the preferred code response above the rejected one, pointwise and reference-free; all 8,866 pairs, pair-weighted over the five criteria). Source: arxiv.org. Saturation forecast: Estimated already saturated. 51 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Skywork-Reward-V2-Qwen3-8B | 79.97 |
| 2 | Llama-3.1-Tulu-3-70B-SFT-RM-RB2 | 78.96 |
| 3 | Llama 3.1 70B Instruct RM RB2 | 78.23 |
| 4 | Llama 3.1 8B Base RM RB2 | 75.77 |
| 5 | LDL-Reward-Gemma-2-27B-v0.1 | 75.57 |
| 6 | INF-ORM-Llama3.1-70B | 74.84 |
| 7 | FsfairX-LLaMA3-RM-v0.1 | 73.52 |
| 8 | InternLM2-20B-Reward | 73.13 |
| 9 | QRM-Llama3.1-8B-v2 | 71.81 |
| 10 | GRM-llama3-8B-sftreg | 71.81 |
| 11 | ArmoRM-Llama3-8B-v0.1 | 71.76 |
| 12 | Llama-3-OffsetBias-RM-8B | 71.76 |
| 13 | InternLM2-7B-Reward | 71.73 |
| 14 | URM-LLaMa-3.1-8B | 71.62 |
| 15 | AceCodeRM-7B | 71.11 |
Interactive version: theaggregate.ai/benchmark?slug=themis-coderewardbench · How It Works · Data refreshed daily, snapshot 2026-09-26.