Multi-Challenge

Multi-Challenge (by Scale AI) tests instruction following across many conversation turns: instructions from earlier turns must not be forgotten, new ones must be added, and contradictions must be resolved. This exposes practical chat behaviour much better than single-turn tests.

Multi-Challenge measures how precisely a model obeys the instructions it was given. In detail, it checks:

• keeping earlier instructions over multiple turns
• combining new constraints with old ones
• long conversations without drifting away from the rules
• precise compliance exactly when it gets difficult

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 Multi-Challenge benchmark?

Multi-Challenge (by Scale AI) tests instruction following across many conversation turns: instructions from earlier turns must not be forgotten, new ones must be added, and contradictions must be resolved. This exposes practical chat behaviour much better than single-turn tests.

What does Multi-Challenge test?

Multi-Challenge mainly measures how precisely a model obeys the instructions it was given. Concretely, it checks: keeping earlier instructions over multiple turns; combining new constraints with old ones; long conversations without drifting away from the rules.

How is the Multi-Challenge 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 Multi-Challenge 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/multi-challenge/scores) or pull it directly from the Open LLM Radar API with any HTTP client.