WMDP (Bio)
WMDP (Bio) - Weapons of Mass Destruction Prohibition (Biology) benchmark evaluating biological knowledge that could enable threats. Measures preparedness risk.
Category
Safety
Score type
percent
Scale
100%
higher is better
Models tested
3
What is WMDP (Bio)?
WMDP-Bio is a multiple-choice test derived from the WMDP (Weapons of Mass Destruction Proxy) corpus: biology knowledge at the level of dangerous dual-use content. In safety research it serves as a measurable proxy for hazardous knowledge — lower scores in a control setting are the goal, raw scores measure exposure.
What does it test?
WMDP (Bio) measures safe and responsible behaviour. In detail, it checks:
- sensitive biology knowledge, represented as multiple-choice accuracy
- responsible handling of dual-use information
- measurable effect of safety training (unlearning) on this knowledge
- risk assessment for publicly distributed models
How to read the score
The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.
Frequently asked questions
What is the WMDP (Bio) benchmark?
WMDP-Bio is a multiple-choice test derived from the WMDP (Weapons of Mass Destruction Proxy) corpus: biology knowledge at the level of dangerous dual-use content. In safety research it serves as a measurable proxy for hazardous knowledge — lower scores in a control setting are the goal, raw scores measure exposure.
What does WMDP (Bio) test?
WMDP (Bio) mainly measures safe and responsible behaviour. Concretely, it checks: sensitive biology knowledge, represented as multiple-choice accuracy; responsible handling of dual-use information; measurable effect of safety training (unlearning) on this knowledge.
How is the WMDP (Bio) score calculated?
The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.
Where do I get WMDP (Bio) data and scores?
The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/wmdp-bio/scores) or pull it directly from the Open LLM Radar API with any HTTP client.
Benchmark data & downloads
Where to get, inspect and reuse the data behind this benchmark.
Model scores 3 models tested
| # | Model | Score | Date | Source | Verified |
|---|---|---|---|---|---|
| 1 |
Muse-Glimmer-30B
Meta Superintelligence Lab
|
86.5%
|
01.08.2026 | Meta Superintelligence Lab (official model card) | Verified |
| 2 |
Gemma4-31B-it
Google
|
85.9%
|
01.08.2026 | Meta Superintelligence Lab (official model card) | Unverified |
| 3 |
Qwen3.6-27B
Qwen Team (Alibaba Cloud)
|
84.8%
|
01.08.2026 | Meta Superintelligence Lab (official model card) | Unverified |