Scientific Review Evaluation on Scientific Review Human Evaluation Dataset
66.7Actionability Win RateProReviewer
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ProReviewerMethod Model=Qwen3-8B, Opponent=AI-Sci-v2, Opponent Model Scale=397B2026.06 | 66.7 | 14.3 | 85.7 | 0 | 57.1 | 9.5 | 76.2 | 9.5 | |
| ProReviewerMethod Model=Qwen3-8B, Opponent=CycleReviewer, Opponent Model Scale=8B2026.06 | 61.5 | 7.7 | 94.9 | 0 | 82.1 | 0 | 82.1 | 2.6 | |
| ProReviewerMethod Model=Qwen3-8B, Opponent=AgentReview, Opponent Model Scale=397B2026.06 | 57.1 | 14.3 | 76.2 | 9.5 | 61.9 | 14.3 | 71.4 | 9.5 | |
| ProReviewerMethod Model=Qwen3-8B, Opponent=DeepReview, Opponent Model Scale=8B2026.06 | 51.3 | 23.1 | 74.4 | 7.7 | 61.5 | 15.4 | 69.2 | 17.9 |