List your company

Formal Theorem Proving on STP In-domain

2,604.24Average Token CostSegment-level
2,221.64764,804.14637,386.6459,969.1437May 12, 2026

Evaluation results

MethodLinks
Segment-levelTraining data=STP2026.052,604.2441.55—
Whole-proof-segTraining data=STP2026.052,906.4150.52—
Step-levelTraining data=STP2026.054,768.5173.5—
Whole-proofTraining data=STP2026.0512,169.0542.67—
Segment-levelTraining data=STP, Base Model=Qwen2.5-Math-7B, Supervision granularity=Segment-level2026.05——97.32
Step-levelTraining data=STP, Base Model=Qwen2.5-Math-7B, Supervision granularity=Step-level2026.05——97.8
Whole-proofTraining data=STP, Base Model=Qwen2.5-Math-7B, Supervision granularity=Whole-proof2026.05——98.12
Whole-proof-segTraining data=STP, Base Model=Qwen2.5-Math-7B, Evaluation protocol=Segment-style search interface2026.05——95.72