List your company

Theory of Mind reasoning on OpenToM narrative

100Location Accuracy (Cg, First)SFT
46.9660.7374.588.27Jun 8, 2026

Evaluation results

MethodLinks
SFTBackbone=Qwen-2.5-3B2026.0610081925688764993.337177.43—
Thinking RFTBackbone=Qwen-2.5-3B2026.069988946791855794.6780835.57
SFTBackbone=Qwen-2.5-7B2026.06998596689079659577.3383.14—
Thinking RFTBackbone=Qwen-2.5-7B2026.06989197859087769587.6789.146
No-Thinking RFTBackbone=Qwen-2.5-7B2026.069784937188823692.677978.714.43
No-Thinking RFTBackbone=Qwen-2.5-3B2026.068150816356553572.675660.1417.29
AutoToMBackbone=Qwen-2.5-7B2026.06576460486660436157.3356.8626.28
Zero-shotBackbone=Qwen-2.5-3B2026.06515028155348394437.6740.5736.86
SimToMBackbone=Qwen-2.5-7B2026.065053444470404754.6745.6749.7133.43
Zero-shotBackbone=Qwen-2.5-7B2026.064951483355533650.6745.6746.4336.71