Mathematical Reasoning on MATH (test) (Accuracy + Verifier Calls)
55.2AccuracyOurs (gates + hybrid + state-k)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ours (gates + hybrid + state-k)configuration=Full2026.02 | 55.2 | 44.8 | — | |
| DR+DADepth=32, 0-SHOT=true2026.01 | 54.5 | — | 2.6 | |
| Beam Searchb=4, N=642026.02 | 51.8 | 64 | — | |
| Gates + Dtypek=fixed2026.02 | 51.8 | 54 | — | |
| DRDepth=32, 0-SHOT=true2026.01 | 51.3 | — | 2.1 | |
| DR+DADepth=16, 0-SHOT=true2026.01 | 50.7 | — | 3.5 | |
| LADepth=32, 0-SHOT=true2026.01 | 48.2 | — | 1 | |
| DRDepth=16, 0-SHOT=true2026.01 | 47.8 | — | 2.8 | |
| Gates-onlygating=active2026.02 | 47.6 | 58 | — | |
| Verify-allgating=none2026.02 | 45 | 64 | — | |
| Majority VoteN=642026.02 | 44.6 | 64 | — | |
| Best-of-NN=642026.02 | 42.4 | 64 | — | |
| LADepth=16, 0-SHOT=true2026.01 | 40.9 | — | 1 | |
| 0-shot CoT2026.02 | 30.6 | — | — |