BioMysteryBench

A study of 503 profession-specific AGENTS.md profiles found that loading full scientific-agent profiles by default did not improve accuracy and cost 2.2 to 4.5 times more per successful call, accordin…

Independent evaluator Vals found that Google's Gemini 3.8 Flash scored 71.7% on BioMysteryBench's human-solvable tasks and 21.6% on its hard tasks in production runs, versus the 88.8% and 56.5% Google…

Anthropic researchers developed BioMysteryBench, a new bioinformatics benchmark that tests Claude's ability to analyze real-world biological datasets and solve open-ended research problems. The benchm…