benchmarks.bio

Biosecurity · Capabilities

BioSecBench-Surveillance results

Biosecurity surveillance tasks that measure an agent's ability to analyze biological threat signals.

Published score: Current V1 evaluation set (102 evaluations) · short horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for BioSecBench-Surveillance. Biosecurity surveillance tasks that measure an agent's ability to analyze biological threat signals. Current V1 covers 102 evaluations and 42 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5.5* (Pi, LSVP, 60.8%); Gemini 4 Argon (Pi, 59.8%); Claude Opus 5.5 (Claude Code, 58.5%). The highest-ranked published run without an LSVP exemption is Gemini 4 Argon (Pi, 59.8%). 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-Surveillance scores
RankModelHarnessProviderScoreVariant
1Claude Opus 5.5*PiAnthropic60.8%LSVP
2Gemini 4 ArgonPiGoogle59.8%
3Claude Opus 5.5Claude CodeAnthropic58.5%
4Claude Opus 5.5*Claude CodeAnthropic57.8%LSVP
5Claude Opus 5.5PiAnthropic55.8%
6Claude Sonnet 5.5*Claude CodeAnthropic54.6%LSVP
7Claude Sonnet 5.5*PiAnthropic54.6%LSVP
8Claude Opus 5PiAnthropic51.5%
9DeepSeek V4.1 FlashPiDeepSeek51.5%
10GPT-6.1 SolPiOpenAI50.8%
11GPT-6.1 SolCodexOpenAI49.4%
12Claude Fable 5.1*PiAnthropic49.0%LSVP
13GPT-6 AstraPiOpenAI48.5%
14Grok 4.6Grok BuildSpaceXAI48.0%
15Grok 4.7PiSpaceXAI48.0%
16GPT-6 SolPiOpenAI47.7%
17Grok 4.6PiSpaceXAI47.5%
18GPT-5.6 SolCodexOpenAI47.3%
19GPT-5.5PiOpenAI47.1%
20GPT-5.6 SolPiOpenAI47.1%
21Claude Opus 4.8PiAnthropic46.9%
22Claude Opus 4.7PiAnthropic45.8%
23Gemini 3.8 FlashPiGoogle44.1%
24Claude Opus 5Claude CodeAnthropic43.6%
25GPT-5.6 LunaCodexOpenAI43.2%
26Claude Opus 4.7Claude CodeAnthropic41.9%
27Gemini 3.7 FlashPiGoogle41.5%
28GPT-6 LunaPiOpenAI41.5%
29GPT-6.1 TerraPiOpenAI39.9%
30Claude Sonnet 5PiAnthropic39.0%
31Kimi K3PiMoonshot38.6%
32Claude Opus 4.8Claude CodeAnthropic37.3%
33GLM 5.3PiZ.ai36.6%
34GPT-6 AstraCodexOpenAI35.2%
35GPT-5.6 LunaPiOpenAI35.1%
36GPT-6.1 TerraCodexOpenAI34.6%
37Claude Sonnet 4.6Claude CodeAnthropic34.5%
38Mistral Large 4.0PiMistral34.0%
39Claude Sonnet 4.6PiAnthropic32.3%
40Claude Sonnet 5Claude CodeAnthropic30.2%
41Nemotron 3 Ultra 550B A55BPiNVIDIA21.2%
42Nemotron 3.5 LightningPiNVIDIA14.1%

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.