List your company

STEM Reasoning on GPQA Diamond (Accuracy, Token Count)

61.62AccuracyCOPT
56.36857.731559.09560.4585May 19, 2026

Evaluation results

MethodLinks
COPTBackbone=Qwen3-8B, Reasoning Effort=High (Peak Accuracy)2026.0561.626,851
CoTBackbone=Qwen3-8B2026.0559.68,123
CoT (Greedy)Backbone=Qwen3-8B2026.0556.577,909