List your company

Truncated CoT Answering on MuSR

33.6AOCMAT-Steer
30.27231.1363232.864Feb 18, 2026

Evaluation results

MethodLinks
MAT-SteerBackbone=Qwen3-14B2026.0233.6
OriginalBackbone=Qwen3-14B2026.0233.2
Faithfulness OnlyBackbone=Qwen3-14B2026.0233
Balanced RewardsBackbone=Qwen3-14B2026.0232.8
REMULBackbone=Qwen3-14B2026.0232.5
Correctness OnlyBackbone=Qwen3-14B2026.0231.7
Hint OptimizedBackbone=Qwen3-14B2026.0230.4