Scientific Reasoning on GPQA-D (test)
68.7AccuracyThink
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ThinkBase Model=Qwen3-4B-Thinking-25072026.02 | 68.7 | 9,041 | 325.8 | |
| SpecExitBase Model=Qwen3-4B-Thinking-25072026.02 | 68.7 | 7,011 | 137 | |
| EAGLE3Base Model=Qwen3-4B-Thinking-25072026.02 | 67.7 | 8,975 | 212.2 | |
| NoThink*Base Model=Qwen3-4B-Thinking-25072026.02 | 67.2 | 8,833 | 276.8 | |
| DEERBase Model=Qwen3-4B-Thinking-25072026.02 | 67.2 | 9,053 | 505.2 | |
| SpecExitBase Model=DeepSeek-R1-Distill-Llama-8B2026.02 | 46 | 6,849 | 307.5 | |
| EAGLE3Base Model=DeepSeek-R1-Distill-Llama-8B2026.02 | 43.9 | 8,749 | 420.1 | |
| VanillaBase Model=DeepSeek-R1-Distill-Llama-8B2026.02 | 43.6 | 8,857 | 574 | |
| DEERBase Model=DeepSeek-R1-Distill-Llama-8B2026.02 | 40.9 | 8,492 | 521.5 | |
| NoThinkBase Model=DeepSeek-R1-Distill-Llama-8B2026.02 | 26.8 | 1,200 | 166.6 |