Themis-CodeRewardBench - Readability and Maintainability: 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; Readability and Maintainability criterion, 1,499 pairs: code-style commits and NoFunEval Maintain). Source: arxiv.org. Saturation forecast: Estimated already saturated. 51 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Skywork-Reward-V2-Qwen3-8B | 75.05 |
| 2 | Llama-3.1-Tulu-3-70B-SFT-RM-RB2 | 73.58 |
| 3 | Llama 3.1 70B Instruct RM RB2 | 71.25 |
| 4 | UltraRM-13B | 71.18 |
| 5 | Llama 3.1 8B Base RM RB2 | 70.78 |
| 6 | INF-ORM-Llama3.1-70B | 68.18 |
| 7 | InternLM2-7B-Reward | 67.38 |
| 8 | FsfairX-LLaMA3-RM-v0.1 | 67.31 |
| 9 | LDL-Reward-Gemma-2-27B-v0.1 | 67.31 |
| 10 | GRM-llama3-8B-sftreg | 66.11 |
| 11 | ArmoRM-Llama3-8B-v0.1 | 64.71 |
| 12 | internlm2-1.8B-reward | 64.71 |
| 13 | Starling-RM-34B | 64.58 |
| 14 | InternLM2-20B-Reward | 64.11 |
| 15 | Llama-3-OffsetBias-RM-8B | 63.91 |
Interactive version: theaggregate.ai/benchmark?slug=themis-coderewardbench-readability-and-maintainability · How It Works · Data refreshed daily, snapshot 2026-09-26.