List your company

Math Reasoning on MATH 500 (multi@5 accuracy)

58.2Multi@5 AccuracyDRIFT-5turn
37.442.848.253.6May 29, 2026

Evaluation results

MethodLinks
DRIFT-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Ours, Max turns=52026.0558.2
UFO-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=52026.0556.4
SCoRe-2turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=22026.0555.8
SFT-5turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=52026.0553.6
PPOBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.0551.2
SFTBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.0550.8
STaR-2turnBackbone=Qwen2.5-3B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=22026.0550.4
DRIFT-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Ours, Max turns=52026.0548.2
UFO-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=52026.0546.4
SCoRe-2turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Online), Max turns=22026.0545
SFT-5turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=52026.0544
STaR-2turnBackbone=Llama3.1-8B-Instruct, Protocol=Multi-turn baselines (Offline), Max turns=22026.0543.2
SFTBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.0542
PPOBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.0541.6
BaseBackbone=Qwen2.5-3B-Instruct, Protocol=Single-turn baselines2026.0540.2
BaseBackbone=Llama3.1-8B-Instruct, Protocol=Single-turn baselines2026.0538.2