LAB-Bench LitQA2 (SecureBio)
Science
SecureBio's own runs of LAB-Bench LitQA2, 199 public multiple-choice questions on recent scientific literature, without retrieval or internet access. Evaluated in Inspect with zero-shot prompting, usually over 10 runs. SecureBio doesn't state how refusals enter accuracy. Refusals reach 17.6% here, and on other tests of the dashboard printed accuracy exceeds what counting refusals as wrong allows, so accuracy appears to cover answered questions only and each model's denominator differs. These scores aren't comparable with other runs of the test, which is why this is a diagnostic.
Benchmark source: Securebio