List your company

Adversarial Game Playing on Don’t Say It

63.02GPT-5.1 PerformanceDEPT
3.979219.307134.63549.9629May 9, 2026

Evaluation results

MethodLinks
DEPTBackbone=Qwen3-8B-Base2026.0563.0252.7347.9254.56
DEPTBackbone=Qwen3-4B-Base2026.0556.6454.9558.5956.73
SPAGBackbone=Qwen3-8B-Base2026.0554.148.4426.4141.21
MARSBackbone=Qwen3-4B-Base2026.0552.4742.1927.9940.89
MARSBackbone=Qwen3-8B-Base2026.0551.9541.5428.3940.62
SPIRALBackbone=Qwen3-4B-Base2026.0548.3142.0647.2745.88
GRPOBackbone=Qwen3-4B-Base2026.0542.1941.9341.9342.01
SPIRALBackbone=Qwen3-8B-Base2026.0541.1529.6942.8437.89
SPAGBackbone=Qwen3-4B-Base2026.0537.1126.8214.5826.17
GRPOBackbone=Qwen3-8B-Base2026.0533.4640.2230.6434.77
VANILLABackbone=Qwen3-8B-Base2026.0523.2416.415.8615.17
VANILLABackbone=Qwen3-4B-Base2026.056.252.471.433.39