Scientific Reasoning on GPQA Main (Accuracy, Tokens)
20.8AccuracyStatic
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| StaticBudget=40962026.04 | 20.8 | 10,994 | |
| TABBudget (B)=10k2026.04 | 19.4 | 6,415 | |
| LLM-Judge Multi-TurnSelection Strategy=Multi-Turn2026.04 | 18.1 | 5,712 | |
| StaticBudget=20482026.04 | 17.4 | 5,910 | |
| LLM-Judge IndividualSelection Strategy=Individual2026.04 | 17.4 | 5,940 | |
| TABBudget (B)=8k2026.04 | 17.2 | 4,843 | |
| StaticBudget=10242026.04 | 16.7 | 3,929 | |
| TABBudget (B)=5k2026.04 | 16.5 | 2,971 | |
| TABBudget (B)=3k2026.04 | 15.6 | 2,258 | |
| StaticBudget=5122026.04 | 14.7 | 2,544 | |
| StaticBudget=2562026.04 | 14.3 | 1,822 |