benchmarks.bio

Biosecurity · Capabilities

BioSecBench-Function results

Biosecurity function-analysis tasks measuring capabilities on biologically sensitive questions.

Published score: Current V1 evaluation set (111 evaluations) · short horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for BioSecBench-Function. Biosecurity function-analysis tasks measuring capabilities on biologically sensitive questions. Current V1 covers 111 evaluations and 42 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5.5 (Pi, 73.6%); Claude Opus 5.5 (Claude Code, 61.5%); Claude Opus 5.5* (Pi, LSVP, 56.1%). 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-Function scores
RankModelHarnessProviderScoreVariant
1Claude Opus 5.5PiAnthropic73.6%
2Claude Opus 5.5Claude CodeAnthropic61.5%
3Claude Opus 5.5*PiAnthropic56.1%LSVP
4Claude Opus 5.5*Claude CodeAnthropic54.2%LSVP
5Claude Sonnet 5.5Claude CodeAnthropic53.5%
6Claude Sonnet 5.5PiAnthropic51.7%
7Claude Fable 5.1*PiAnthropic49.0%LSVP
8GPT-6 AstraPiOpenAI48.6%
9GPT-6.1 SolPiOpenAI46.0%
10GPT-6 AstraCodexOpenAI46.0%
11Claude Opus 5Claude CodeAnthropic44.8%
12GPT-6.1 SolCodexOpenAI44.6%
13Gemini 4 ArgonPiGoogle44.4%
14Grok 4.6PiSpaceXAI44.0%
15Grok 4.7PiSpaceXAI43.3%
16Claude Opus 4.8PiAnthropic43.1%
17Claude Opus 5PiAnthropic42.5%
18DeepSeek V4.1 FlashPiDeepSeek40.8%
19Grok 4.6Grok BuildSpaceXAI39.9%
20GPT-6 SolPiOpenAI39.3%
21GPT-6.1 TerraPiOpenAI39.0%
22Claude Opus 4.8Claude CodeAnthropic39.0%
23GPT-6.1 TerraCodexOpenAI38.3%
24Gemini 3.8 FlashPiGoogle36.0%
25GPT-5.6 SolCodexOpenAI35.5%
26GPT-5.6 SolPiOpenAI34.4%
27Claude Sonnet 5Claude CodeAnthropic33.8%
28Gemini 3.7 FlashPiGoogle33.7%
29Claude Sonnet 5PiAnthropic33.1%
30Kimi K3PiMoonshot32.7%
31GLM 5.3PiZ.ai31.8%
32Claude Opus 4.7PiAnthropic30.8%
33Claude Opus 4.7Claude CodeAnthropic30.3%
34Mistral Large 4.0PiMistral28.8%
35GPT-5.5PiOpenAI28.3%
36GPT-5.6 LunaPiOpenAI27.6%
37Claude Sonnet 4.6Claude CodeAnthropic27.3%
38GPT-6 LunaPiOpenAI27.3%
39GPT-5.6 LunaCodexOpenAI26.3%
40Claude Sonnet 4.6PiAnthropic25.5%
41Nemotron 3 Ultra 550B A55BPiNVIDIA21.6%
42Nemotron 3.5 LightningPiNVIDIA13.8%

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.