Themis-CodeRewardBench - Memory Efficiency: 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; Memory Efficiency criterion, 289 pairs: memory-improving commits and NoFunEval Memory). Source: arxiv.org. Saturation forecast: Estimated already saturated. 51 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Skywork-Reward-V2-Qwen3-8B | 71.63 |
| 2 | Llama 3.1 8B Base RM RB2 | 69.9 |
| 3 | FsfairX-LLaMA3-RM-v0.1 | 67.82 |
| 4 | Llama-3.1-Tulu-3-70B-SFT-RM-RB2 | 67.13 |
| 5 | QRM-Llama3.1-8B-v2 | 66.78 |
| 6 | UltraRM-13B | 66.44 |
| 7 | InternLM2-20B-Reward | 65.74 |
| 8 | URM-LLaMa-3.1-8B | 65.05 |
| 9 | Llama 3.1 70B Instruct RM RB2 | 64.71 |
| 10 | Starling-RM-34B | 63.67 |
| 11 | INF-ORM-Llama3.1-70B | 62.63 |
| 12 | LDL-Reward-Gemma-2-27B-v0.1 | 62.63 |
| 13 | InternLM2-7B-Reward | 62.28 |
| 14 | internlm2-1.8B-reward | 62.28 |
| 15 | AceCodeRM-7B | 60.55 |
Interactive version: theaggregate.ai/benchmark?slug=themis-coderewardbench-memory-efficiency · How It Works · Data refreshed daily, snapshot 2026-09-26.