Benchmark
LabBench
An evaluation benchmark close to wet-lab reasoning that requires models to understand experimental design, interpret results and follow causal reasoning chains to perform multi-step inference.
Timeline (last 90 days)
Oct 13
On RareBench, Gemini 4 Argon ranked first with a 57% assisted diagnosis rate, and scored 52.2% on LabBench wet-lab reasoning.
Unverified50%