List your company

Mathematical Reasoning on Math Benchmarks (MATH, MATH500, ThmQA)

67.6MATH multi@5 AccuracyDRIFT-5turn
61.98463.44264.966.358May 29, 2026

Evaluation results

MethodLinks
DRIFT-5turnModel=Qwen2.5-7B-Instruct, Maximum budget=5 turns2026.0567.668.638.558.2
UFO-5turnModel=Qwen2.5-7B-Instruct, Maximum budget=5 turns2026.0566.366.238.657
SFT-5turnModel=Qwen2.5-7B-Instruct, Maximum budget=5 turns2026.0564.264.437.255.3
BaseModel=Qwen2.5-7B-Instruct, Maximum budget=5 turns2026.0562.26437.154.4