BioSecBench-Surveillance V0 (102 tasks)
Science
Original BioSecBench-Surveillance cohort, formerly called V1 and now V0. Macro-mean pass rate over 102 genomic surveillance tasks, distinct from the earlier 100-task release and clean reruns.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 51.5 |
| GPT-6 Astra | 48.5 |
| Grok 4.6 | 48.0 |
| Gemini 3.8 Flash | 44.1 |
| Gemini 3.7 Flash | 41.5 |