ResearchBenchmarksExpert-level Science Question Answering on GPQA diamond (Acc, #Tokens)Follow47.47AccuracyMUR32.244436.197240.1544.1028Jul 20, 2025Evaluation ResultsMethodMethodLinksAccuracyOutput Tokens CountMURBackbone=Qwen3-8B, Sea...Backbone=Qwen3-8B, Search Strategy=phi-Decoding2025.0747.472,359MURBackbone=Qwen3-4B, Sea...Backbone=Qwen3-4B, Search Strategy=phi-Decoding2025.0741.92944MURBackbone=Qwen3-1.7B, S...Backbone=Qwen3-1.7B, Search Strategy=Guided search2025.0732.832,005