STEM Reasoning on GPQA Diamond (Accuracy, Token Count)
61.62AccuracyCOPT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| COPTBackbone=Qwen3-8B, Reasoning Effort=High (Peak Accuracy)2026.05 | 61.62 | 6,851 | |
| CoTBackbone=Qwen3-8B2026.05 | 59.6 | 8,123 | |
| CoT (Greedy)Backbone=Qwen3-8B2026.05 | 56.57 | 7,909 |