Skywork-Reward-Llama-3.1-8B-v0.2: benchmark results

Provider: Skywork. Access: Open.

Unified ELO 1808 ± 52, rank #142 of 2656 rated models, from 11 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
RewardBench Safety94.22Accuracy (%)97.5
RewardBench93.13Score (%)96.9
RewardBench Chat Hard88.38Accuracy (%)95.5
RewardBench Focus94.14Score (%)93.6
RewardBench Reasoning96.75Accuracy (%)91.7
RewardBench Ties71.69Score (%)79.5
RewardBench Precise IF40.62Score (%)77.3
RewardBench Factuality69.68Accuracy (%)58.2
RewardBench Chat94.69Accuracy (%)53.1
StoryAlign31Average (self-reported)35.7
RewardBench Math60.11Score (%)33.7

Interactive version: theaggregate.ai/model?slug=skywork-reward-llama-3-1-8b-v0-2 · How It Works · Data refreshed daily, snapshot 2026-09-19.