Scientific Reasoning on GPQA-D (Acc, Tok, CR)
69.5AccuracyVanilla
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VanillaBackbone=Gemini2.5-Flash2026.04 | 69.5 | 4,565 | 100 | |
| DynasorBackbone=Gemini2.5-Flash2026.04 | 69.5 | 4,091 | 89.6 | |
| TRACEBackbone=Gemini2.5-Flash2026.04 | 69.4 | 3,262 | 71.4 | |
| DEERBackbone=Gemini2.5-Flash2026.04 | 69.1 | 3,397 | 74.4 | |
| TALEBackbone=Gemini2.5-Flash2026.04 | 68.2 | 3,597 | 78.8 | |
| NoThinkBackbone=Gemini2.5-Flash2026.04 | 67.3 | 4,365 | 95.6 | |
| VanillaBackbone=Qwen3-8B2026.04 | 59.1 | 9,398 | 100 | |
| TRACEBackbone=Qwen3-8B2026.04 | 56.6 | 5,629 | 59.9 | |
| VanillaBackbone=Qwen3-4B2026.04 | 55.1 | 8,611 | 100 | |
| TRACEBackbone=Qwen3-4B2026.04 | 55 | 5,713 | 67.3 | |
| DynasorBackbone=Qwen3-8B2026.04 | 54.6 | 5,732 | 61 | |
| DEERBackbone=Qwen3-4B2026.04 | 54.5 | 7,042 | 81.8 | |
| DEERBackbone=Qwen3-8B2026.04 | 53.2 | 7,279 | 77.5 | |
| DynasorBackbone=Qwen3-4B2026.04 | 52.1 | 4,814 | 55.9 | |
| TRACEBackbone=R1-Distilled-Llama-8B2026.04 | 51.5 | 4,644 | 75.1 | |
| NoThinkBackbone=Qwen3-8B2026.04 | 51 | 1,538 | 16.4 | |
| VanillaBackbone=R1-Distilled-Llama-8B2026.04 | 51 | 6,184 | 100 | |
| NoThinkBackbone=R1-Distilled-Llama-8B2026.04 | 50.5 | 5,849 | 94.6 | |
| TALEBackbone=Qwen3-8B2026.04 | 50 | 4,245 | 45.2 | |
| NoThinkBackbone=Qwen3-4B2026.04 | 49 | 1,658 | 19.3 | |
| TALEBackbone=Qwen3-4B2026.04 | 47.9 | 5,807 | 67.4 | |
| DynasorBackbone=R1-Distilled-Llama-8B2026.04 | 43.6 | 4,019 | 65 | |
| DEERBackbone=R1-Distilled-Llama-8B2026.04 | 43 | 4,731 | 76.5 | |
| TALEBackbone=R1-Distilled-Llama-8B2026.04 | 37.4 | 4,531 | 73.3 |