WMDP (Chem)

WMDP-Chem is the chemistry part of the WMDP proxy corpus: dangerous dual-use chemistry knowledge in the form of expert multiple-choice questions. It is used to measure how much hazardous knowledge a model carries — and whether safety training can remove it without damaging general chemistry.

WMDP (Chem) measures safe and responsible behaviour. In detail, it checks:

• sensitive chemistry knowledge as a measurable proxy
• effect of safety/alignment training on hazardous content
• retention of benign scientific chemistry ability
• responsible distribution decisions for released models

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the WMDP (Chem) benchmark?

WMDP-Chem is the chemistry part of the WMDP proxy corpus: dangerous dual-use chemistry knowledge in the form of expert multiple-choice questions. It is used to measure how much hazardous knowledge a model carries — and whether safety training can remove it without damaging general chemistry.

What does WMDP (Chem) test?

WMDP (Chem) mainly measures safe and responsible behaviour. Concretely, it checks: sensitive chemistry knowledge as a measurable proxy; effect of safety/alignment training on hazardous content; retention of benign scientific chemistry ability.

How is the WMDP (Chem) 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 (Chem) 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-chem/scores) or pull it directly from the Open LLM Radar API with any HTTP client.