List your company

Compositional Reasoning on GPQA

48.9AccuracyRCE
23.10829.80436.543.196Feb 17, 2026

Evaluation results

MethodLinks
RCEModel=Qwen-14B2026.0248.9
RCEModel=Llama-3-8B2026.0243.1
RCEModel=Mistral-7B2026.0241.4
BaseModel=Qwen-14B2026.0236.7
DisCOModel=Mistral-7B2026.0234.2
GRPOModel=Mistral-7B2026.0232.4
ToTModel=Mistral-7B2026.0231
SCModel=Mistral-7B, Samples (n)=162026.0230.3
CoTModel=Mistral-7B2026.0228.5
BaseModel=Llama-3-8B2026.0227.3
BaseModel=Mistral-7B2026.0224.1