Themis-CodeRewardBench - Security Hardness: 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; Security Hardness criterion, 991 pairs: vulnerability-fix commits, CodePrefBench Security, Vul4J, SecBench and NoFunEval Security). Source: arxiv.org. Saturation forecast: Estimated already saturated. 51 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Skywork-Reward-V2-Qwen3-8B | 74.97 |
| 2 | Llama-3.1-Tulu-3-70B-SFT-RM-RB2 | 73.76 |
| 3 | Llama 3.1 70B Instruct RM RB2 | 72.96 |
| 4 | UltraRM-13B | 72.45 |
| 5 | Llama 3.1 8B Base RM RB2 | 71.24 |
| 6 | Starling-RM-34B | 68.72 |
| 7 | GRM-llama3-8B-sftreg | 67 |
| 8 | FsfairX-LLaMA3-RM-v0.1 | 66.6 |
| 9 | INF-ORM-Llama3.1-70B | 66.6 |
| 10 | InternLM2-20B-Reward | 66.26 |
| 11 | Llama-3-OffsetBias-RM-8B | 65.29 |
| 12 | InternLM2-7B-Reward | 63.57 |
| 13 | LDL-Reward-Gemma-2-27B-v0.1 | 63.17 |
| 14 | Eurus-RM-7B | 63.07 |
| 15 | QRM-Llama3.1-8B-v2 | 62.36 |
Interactive version: theaggregate.ai/benchmark?slug=themis-coderewardbench-security-hardness · How It Works · Data refreshed daily, snapshot 2026-09-26.