Long-form Biological Risk Questions
OpenAI deep research 0% leads since Feb 2025 Only in system cards
Models answer long-form questions across five biothreat creation stages, graded against expert rubrics. Scored as pass@1 per stage.
OpenAI Preparedness evaluation. Cards report stage scores with no overall number.
Made by OpenAI with Gryphon Scientific Released Sep 2024 Hazardous knowledge Tasks kept private Last in a system card Nov 2025
Last model added Nov 2025
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
3
Models scored in lab cards
14
System cards report it, from 1 lab
3 models share the top score of 0% among models released by Nov 2025
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in Nov 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 1 OpenAI deep research 0%
- 2 GPT-4.5 0%
- 3 GPT-5.1-Codex-Max 0%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | OpenAI deep research | 0% | Feb 2025 | Model card | |
| 2 | GPT-4.5 | 0% | Feb 2025 | Model card | |
| 3 | GPT-5.1-Codex-Max | 0% | Nov 2025 | Model card |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
OpenAI
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Reported by the lab in its model card
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
14 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 18 Nov 2025 | GPT-5.1-Codex-Max system card | 0 | Long-form biorisk questions, Autograded Gryphon Free Response, gpt-5.1-codex-max 0% on all 5 stages, text says it refused 100% of prompts (Figure 1, 5.1.1.1) | Bio and chem risk |
| 7 Aug 2025 | GPT-5 system card | Long-form Biological Risk Questions (Autograded Gryphon Free Response), 5 stages, no overall score (Figure 9) | Bio and chem risk | |
| 7 Aug 2025 | GPT-5 mini system card | Long-form Biological Risk Questions (Autograded Gryphon Free Response), gpt-5-thinking-mini browsing launch candidate, 5 stages, no overall score (Figure 9) | Bio and chem risk | |
| 5 Aug 2025 | gpt-oss-20b model card | Long-form Biological Risk Questions (Autograded Gryphon Free Response), gpt-oss 20b browsing launch candidate, 5 stages, no overall score (Figure 5) | Bio and chem risk | |
| 5 Aug 2025 | gpt-oss-120b model card | Long-form Biological Risk Questions (Autograded Gryphon Free Response), gpt-oss 120b browsing launch candidate, 5 stages, no overall score (Figure 5) | Bio and chem risk | |
| 17 Jul 2025 | ChatGPT agent system card | Long-form Biological Risk Questions, Autograded Gryphon Free Response, chart shows only the ChatGPT Agent browsing helpful-only variant, no released-model value (Figure 1, 5.1.1.1) | Bio and chem risk | |
| 16 Apr 2025 | o4-mini system card | Long-form biorisk questions, autograded Gryphon free response, 5 stages for helpful-only variants only, no overall score (Figure 3) | Bio and chem risk | |
| 16 Apr 2025 | o3 system card | Long-form biorisk questions, autograded Gryphon free response, 5 stages for helpful-only variants only, no overall score (Figure 3) | Bio and chem risk | |
| 27 Feb 2025 | GPT-4.5 system card | 0 | Long-form Biological Risk Questions, post-mitigation scores 0% on all five stages due to refusals (4.3.1) | Bio and chem risk |
| 25 Feb 2025 | OpenAI deep research system card | 0 | Long-form Biological Risk Questions, chart plots pre-mitigation models only, text says post-mitigation models scored 0% across all physical success steps (3.4.4) | Bio and chem risk |
| 31 Jan 2025 | o3-mini system card | Long-form Biological Risk Questions, chart for pre-mitigation stages, text says post-mitigation models reliably refuse, no post-mitigation number (5.4.1) | Bio and chem risk | |
| 5 Dec 2024 | o1 system card | Biothreat Information Long-Form, accuracy for 5 stages, o1 (pre-mitigation) only, no overall score (5.5.1 chart) | Bio and chem risk | |
| 12 Sep 2024 | o1-preview system card | Biothreat Information Long-Form, accuracy for 5 stages, o1-preview (pre-mitigation) only, chart without printed numbers (4.3.1) | Bio and chem risk | |
| 12 Sep 2024 | o1-mini system card | Biothreat Information Long-Form, accuracy for 5 stages, o1-mini (pre-mitigation) only, chart without printed numbers (4.3.1) | Bio and chem risk |
More in Biology and chemistry
- SUPERChem In use Models solve 500 expert-written chemistry reasoning problems, offered in multimodal and text-only forms. Expert solution paths also allow a reasoning path score.
Tencent Hy4 preview 66.4% - GeneBench Pro In use Agents analyze messy simulated genomics and biology datasets to estimate a decision-relevant quantity across 129 problems. Scored automatically against known answers. GPT-6 Astra 37.1%
- LifeSciBench In use Models complete 750 expert-written life science research tasks, often with attached figures, tables, or sequence files. Graded against expert rubrics as score and pass rate. GPT-6 Astra 60.3%
- MolecularIQ In use Models answer questions about molecular graphs, such as counting, indexing and constrained generation, and a symbolic checker verifies each answer.
Intern-S2-397B 62.4% - TOMG-Bench In use Models generate molecules from natural language requests in three tasks: molecule editing, molecule optimization and customized generation. Each output is checked for validity and for meeting the requested properties. Intern-S2-397B 66.8%
- BABE Fading Models reason from experimental results in peer-reviewed biology papers to draw conclusions, with causal and cross-scale inference.
Seed2.0 Lite (0428) 57.9%