Scientific Reasoning on GPQA-D (Acc, Think Tok.)
5,318Thinking TokensDRE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DREBackbone=Qwen3-30B-A3B-Thinking-25072026.06 | 5,318 | 75.51 | |
| DREBackbone=Qwen3-4B-Thinking-25072026.06 | 5,689 | 65.67 | |
| RCPDBackbone=Qwen3-30B-A3B-Thinking-25072026.06 | 6,021 | 74.94 | |
| DEERBackbone=Qwen3-30B-A3B-Thinking-25072026.06 | 6,528 | 75.47 | |
| DREBackbone=Qwen3-8B2026.06 | 6,658 | 60.98 | |
| Raw ModelBackbone=Qwen3-30B-A3B-Thinking-25072026.06 | 6,922 | 76.45 | |
| RCPDBackbone=Qwen3-4B-Thinking-25072026.06 | 7,265 | 66.04 | |
| RCPDBackbone=Qwen3-8B2026.06 | 7,330 | 59.85 | |
| DEERBackbone=Qwen3-4B-Thinking-25072026.06 | 7,914 | 66.72 | |
| Raw ModelBackbone=Qwen3-4B-Thinking-25072026.06 | 8,091 | 65.47 | |
| DEERBackbone=Qwen3-8B2026.06 | 8,138 | 60.98 | |
| Raw ModelBackbone=Qwen3-8B2026.06 | 8,683 | 59.34 |