Skywork-Reward-Gemma-2-27B-v0.2: benchmark results
Skywork's Gemma 2 27B-based reward model, formerly #1 on RewardBench. Provider: Skywork. Released 2024-10-14. Access: Open.
Unified ELO 1608 ± 1, rank #241 of 1392 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| RewardBench Safety | 96.89 | Accuracy (%) | 99.7 |
| RewardBench | 94.26 | Score (%) | 99.1 |
| RewardBench Chat Hard | 89.91 | Accuracy (%) | 97.2 |
| RewardBench Reasoning | 98.07 | Accuracy (%) | 96.2 |
| Open LLM Leaderboard - IFEval | 78.07 | Score | 94.2 |
| RewardBench Ties | 81.82 | Score (%) | 92.6 |
| RewardBench Factuality | 76.74 | Accuracy (%) | 92.3 |
| RewardBench Focus | 91.72 | Score (%) | 89.3 |
| Open LLM Leaderboard - BBH | 48.16 | Score | 88.2 |
| Open LLM Leaderboard - GPQA | 12.53 | Score | 85.8 |
| RewardBench Math | 67.21 | Score (%) | 80.9 |
| Open LLM Leaderboard - MATH Level 5 | 22.73 | Score | 76.1 |
Interactive version: theaggregate.ai/model?slug=skywork-reward-gemma-2-27b-v0-2 · How It Works · Data refreshed daily, snapshot 2026-09-05.