Bioweapons acquisition uplift trial (Anthropic)
Participants draft a bioweapons acquisition plan over two days with or without model access. Score is the uplift ratio of model group over control.
In-house Anthropic human uplift trial graded by Deloitte with a rubric. The tasks are not public.
- Made by
- Anthropic
- Kind
- Hazardous knowledge
- Level
- Expert
- Scored
- Graded by people
- Answers
- Open-ended
- Tasks
- Kept private
- Strongest model tested
- Claude Opus 4 #108 on the capability index
- Last in a system card
- May 2025
- Last model added
- May 2025
Claude Opus 4 leads with 2.53 among models released by May 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 May 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 Claude Opus 4 2.53
- 2 Claude 3.7 Sonnet 2.10
- 3 Claude Sonnet 4 1.70
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4 | 2.53 | May 2025 | Lab model card | |
| 2 | Claude 3.7 Sonnet | 2.10 | Feb 2025 | Lab model card | |
| 3 | Claude Sonnet 4 | 1.70 | May 2025 | Lab model card |
2 models have held number one in labs’ own numbers since 2023
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Labs’ model cards
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
3 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 |
|---|---|---|---|---|
| May 22, 2025 | Claude Sonnet 4 system card | 1.7 | Bioweapons acquisition uplift trial, total uplift 1.70x over internet-only control, earlier snapshot. Sonnet 4 group raw score 42% ± 11%, control 25% ± 13% (7.2.4.1) | Bio and chem risk |
| May 22, 2025 | Claude Opus 4 system card | 2.53 | Bioweapons acquisition uplift trial, total uplift 2.53x over internet-only control. Opus 4 group raw score 63% ± 13%, control 25% ± 13% (7.2.4.1) | Bio and chem risk |
| Feb 24, 2025 | Claude 3.7 Sonnet system card | 2.1 | Bioweapons acquisition uplift trial, within-group uplift ~2.1X with a variant of Claude 3.7 Sonnet. Sepal 50% ± 21% vs control 24% ± 9%, Anthropic 57% ± 20% vs control 27% ± 9% (7.1.4, Figure 15) | Bio and chem risk |
More in Biology and chemistry
- 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-5.6 Sol 31.5%
- 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% - 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-Rosalind-6 61.3%
- MedChemBench (OpenAI) In use Models answer chemistry questions from an OpenAI internal set, grouped with OpenAI's life science research evaluations. OpenAI has not published the task format. GPT-Rosalind-6 49.6%
- ProtocolQA Open-Ended In use Models correct errors in published lab protocols using open-ended questions adapted from LAB-Bench ProtocolQA. Scored as pass@1. GPT-6 Cyber 47.2%
- MolecularIQ In use Models answer questions about molecular graphs, such as counting, indexing and constrained generation, and a symbolic checker verifies each answer. gpt-oss-120b 47.5%