benchmarks.bio

Biosecurity · Capabilities

BioSecBench-RespiratoryVirus-Countermeasures results

Tests whether AI agents can analyze biological data, laboratory results, and patient records to understand how respiratory viruses cause illness, how immunity protects people, and whether vaccines and treatments remain effective as viruses change. These capabilities support research on preventing infections and reducing severe disease. Scores show the share of attempts that reach the correct answer, counting refusals and errors as failures. A higher score means the model is more accurate at these tasks.

Published score: Current V1 evaluation set (64 evaluations) · short horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for BioSecBench-RespiratoryVirus-Countermeasures. Tests whether AI agents can analyze biological data, laboratory results, and patient records to understand how respiratory viruses cause illness, how immunity protects people, and whether vaccines and treatments remain effective as viruses change. These capabilities support research on preventing infections and reducing severe disease. Scores show the share of attempts that reach the correct answer, counting refusals and errors as failures. A higher score means the model is more accurate at these tasks. Current V1 covers 64 evaluations and 30 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Fable 5.2* (Pi, LSVP, 33.3%); Claude Fable 5.2* (Claude Code, LSVP, 32.3%); Gemini 4 Carbon (Pi, 30.2%). The highest-ranked published run without an LSVP exemption is Gemini 4 Carbon (Pi, 30.2%). Scores report benchmark-specific biological task performance on a 0–100 percentage scale. See the linked paper and repository for task construction, grading and uncertainty methodology. Results describe the complete model and agent harness used for each run, and are not comparable across different benchmarks or evaluation versions.

Published BioSecBench-RespiratoryVirus-Countermeasures scores
RankModelHarnessProviderScoreVariant
1Claude Fable 5.2*PiAnthropic33.3%LSVP
2Claude Fable 5.2*Claude CodeAnthropic32.3%LSVP
3Gemini 4 CarbonPiGoogle30.2%
4Claude Opus 5.5*PiAnthropic27.1%LSVP
5Claude Opus 5.5*Claude CodeAnthropic25.5%LSVP
6Claude Sonnet 5.5PiAnthropic18.8%
7Claude Sonnet 5.5*PiAnthropic18.8%LSVP
8Claude Sonnet 5.5*Claude CodeAnthropic18.2%LSVP
9Claude Fable 5.2Claude CodeAnthropic17.7%
10Claude Fable 5.2PiAnthropic16.9%
11Claude Sonnet 5.5Claude CodeAnthropic16.7%
12GPT-6 AstraPiOpenAI13.5%
13Claude Opus 5.5Claude CodeAnthropic13.0%
14Claude Opus 5.5PiAnthropic12.0%
15Claude Opus 5*Claude CodeAnthropic12.0%LSVP
16Claude Opus 5*PiAnthropic12.0%LSVP
17Gemini 4 ArgonPiGoogle12.0%
18GPT-6.1 TerraPiOpenAI11.5%
19Grok 4.7PiSpaceXAI11.5%
20GPT-6.1 SolPiOpenAI10.9%
21GPT-6 AstraCodexOpenAI9.9%
22Grok 4.6PiSpaceXAI9.4%
23DeepSeek V4.1 FlashPiDeepSeek6.3%
24GLM 5.3PiZ.ai6.3%
25Mistral Large 4.0PiMistral4.2%
26Gemini 3.8 FlashPiGoogle2.6%
27GPT-6 LunaPiOpenAI2.1%
28Kimi K3.1 ProPiMoonshot1.6%
29Nemotron 3 Ultra 550B A55BPiNVIDIA1.0%
30Nemotron 3.5 LightningPiNVIDIA0.0%

Methodology and provenance

Scores use a 0–100 percentage scale. See Paper PDF, Original publication, Repository for task construction, grading, aggregation, and uncertainty details.

LSVP: Anthropic's safety-filter-exempt evaluation variant, which produces fewer refusals on biology prompts. LSVP results are reported alongside the model's standard run and are not a replacement for it.