Benchmark

BrowseComp

Hard fact-seeking browsing tasks that need many retrieval steps to answer.

Score history →

Model scores

#ModelScore
1MiniMax M383.5%