List your company

Truncated CoT Answering on ZebraLogicBench (ZLB)

58.7AOCFaithfulness Only
46.94849.99953.0556.101Feb 18, 2026

Evaluation results

MethodLinks
Faithfulness OnlyBackbone=Qwen3-14B2026.0258.7
REMULBackbone=Qwen3-14B2026.0256.6
MAT-SteerBackbone=Qwen3-14B2026.0252.7
OriginalBackbone=Qwen3-14B2026.0252
Balanced RewardsBackbone=Qwen3-14B2026.0251.2
Hint OptimizedBackbone=Qwen3-14B2026.0248.1
Correctness OnlyBackbone=Qwen3-14B2026.0247.4