Scientific Reasoning on Science GPQA Diamond (test)
63.1GPQA Diamond ScoreSelf-Verified Distillation
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Self-Verified DistillationBackbone=Qwen3-8B, Candidate generations (n)=8, Verification strength (v)=52026.05 | 63.1 | 12 | 2.7 | |
| Qwen3-8BModel size=8B2026.05 | 61.1 | 8.7 | — | |
| Self-Verified DistillationBackbone=Qwen3-4B, Candidate generations (n)=8, Verification strength (v)=52026.05 | 60.4 | 10 | 5.6 | |
| Qwen3-4BModel size=4B2026.05 | 50.8 | 8.5 | — | |
| Self-Verified DistillationBackbone=Qwen3-0.6B, Candidate generations (n)=8, Verification strength (v)=52026.05 | 26.4 | 11.4 | 4.4 | |
| Qwen3-0.6BModel size=0.6B2026.05 | 16.7 | 12.3 | — |