# BioSecBench-Structure results Compare current model-and-harness results for BioSecBench-Structure. BioSecBench-Structure tests whether AI agents can correctly reason about 3D structures of pathogen-associated molecules that guide the design of vaccines, antibodies and antiviral drugs. Tasks include identifying the metal ions in a viral enzyme before antiviral design, finding which parts of a viral surface protein an antibody can reach, and confirming whether a candidate drug is really bound. Each pairs real structural data with a specific analysis question and a verifiable answer. Current V1 covers 58 evaluations and 47 ranked configurations. Data updated 2026-11-12. The highest-ranked published configurations are Claude Fable 5.2* (Pi, LSVP, 52.3%); Claude Fable 5.2* (Claude Code, LSVP, 47.1%); Claude Fable 5.2 (Claude Code, 43.7%). The highest-ranked published run without an LSVP exemption is Claude Fable 5.2 (Claude Code, 43.7%). 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. 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. Data updated: 2026-11-12 Artifact generated: 2026-11-12T18:37:33.767Z - Domain: Biosecurity - Horizon: short - Assessment: Capabilities - Published score scope: Current V1 evaluation set (58 evaluations) Scores use a 0–100 percentage scale. Consult the repository and paper for benchmark-specific task construction, grading, aggregation, and uncertainty methodology. Results compare model-and-harness combinations. 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. | Rank | Model | Harness | Provider | Score | Variant | | ---: | --- | --- | --- | ---: | --- | | 1 | Claude Fable 5.2* | Pi | Anthropic | 52.3% | LSVP | | 2 | Claude Fable 5.2* | Claude Code | Anthropic | 47.1% | LSVP | | 3 | Claude Fable 5.2 | Claude Code | Anthropic | 43.7% | | | 4 | Claude Fable 5.2 | Pi | Anthropic | 41.4% | | | 5 | Claude Opus 5.5* | Pi | Anthropic | 39.7% | LSVP | | 6 | Claude Opus 5.5* | Claude Code | Anthropic | 35.1% | LSVP | | 7 | Claude Opus 5.5 | Claude Code | Anthropic | 31.0% | | | 8 | Claude Opus 5.5 | Pi | Anthropic | 29.3% | | | 9 | DeepSeek V4.1 Flash | Pi | DeepSeek | 27.6% | | | 10 | Claude Sonnet 5.5* | Pi | Anthropic | 27.0% | LSVP | | 11 | Claude Opus 5* | Claude Code | Anthropic | 25.9% | LSVP | | 12 | Claude Sonnet 5.5* | Claude Code | Anthropic | 25.9% | LSVP | | 13 | GPT-6.1 Terra | Pi | OpenAI | 25.9% | | | 14 | Mistral Large 4.0 | Pi | Mistral | 23.0% | | | 15 | Claude Sonnet 5.5 | Claude Code | Anthropic | 21.8% | | | 16 | Claude Sonnet 5.5 | Pi | Anthropic | 21.8% | | | 17 | Claude Sonnet 5* | Claude Code | Anthropic | 21.8% | LSVP | | 18 | Kimi K3 | Pi | Moonshot | 21.8% | | | 19 | GPT-6 Astra | Codex | OpenAI | 21.3% | | | 20 | GPT-6.1 Sol | Pi | OpenAI | 21.3% | | | 21 | GPT-6.1 Terra | Codex | OpenAI | 21.3% | | | 22 | Grok 4.7 | Pi | SpaceXAI | 21.3% | | | 23 | Claude Opus 5 | Claude Code | Anthropic | 20.7% | | | 24 | Claude Sonnet 5 | Claude Code | Anthropic | 20.7% | | | 25 | GPT-5.6 Sol | Pi | OpenAI | 20.7% | | | 26 | GPT-5.6 Luna | Pi | OpenAI | 20.1% | | | 27 | Grok 4.7 | Grok Build | SpaceXAI | 20.1% | | | 28 | GPT-6 Astra | Pi | OpenAI | 19.5% | | | 29 | Claude Sonnet 5 | Pi | Anthropic | 19.0% | | | 30 | Claude Sonnet 5* | Pi | Anthropic | 19.0% | LSVP | | 31 | GLM 5.3 | Pi | Z.ai | 19.0% | | | 32 | GPT-5.6 Sol | Codex | OpenAI | 19.0% | | | 33 | GPT-5.6 Luna | Codex | OpenAI | 18.4% | | | 34 | Claude Opus 5* | Pi | Anthropic | 17.8% | LSVP | | 35 | GPT-6.1 Sol | Codex | OpenAI | 17.2% | | | 36 | Claude Opus 5 | Pi | Anthropic | 15.5% | | | 37 | Grok 4.6 | Pi | SpaceXAI | 15.5% | | | 38 | Gemini 4 Argon | Pi | Google | 14.9% | | | 39 | Grok 4.6 | Grok Build | SpaceXAI | 14.9% | | | 40 | Claude Fable 5.1* | Pi | Anthropic | 13.8% | LSVP | | 41 | GPT-6 Sol | Pi | OpenAI | 13.8% | | | 42 | Claude Fable 5.1* | Claude Code | Anthropic | 13.2% | LSVP | | 43 | GPT-6 Luna | Pi | OpenAI | 13.2% | | | 44 | GPT-6 Luna | Codex | OpenAI | 12.6% | | | 45 | GPT-6 Sol | Codex | OpenAI | 10.9% | | | 46 | Claude Fable 5.1 | Pi | Anthropic | 6.3% | | | 47 | Claude Fable 5.1 | Claude Code | Anthropic | 5.8% | | ## LSVP access Anthropic's Life Sciences Verification Program accepts interest from verified life sciences organizations using Claude Team, Enterprise or the first-party API. Eligibility, supported models and grant types are determined by Anthropic. Program access does not establish equivalence to every benchmark evaluation variant. Guidance checked 2026-11-10. [Anthropic eligibility guidance](https://support.claude.com/en/articles/16975617-what-is-the-life-sciences-verification-program) · [Register organizational interest](https://claude.com/form/life-sciences-verification-program). ## Sources and downloads - [Paper PDF](/papers/structurebench.pdf) - [Original publication](https://latch.bio/structurebench) - [Repository](https://github.com/latchbio/biosecbench-structure) - [Interactive benchmark](https://benchmarks.bio/biosec-structure) - [JSON](https://benchmarks.bio/benchmarks/biosecbench-structure/latest.json) - [CSV](https://benchmarks.bio/benchmarks/biosecbench-structure/latest.csv) - [Current V1 cross-benchmark leaderboard](https://benchmarks.bio/results/v1/)