Inspect

Inspect is an open source framework for evaluating large language models, developed by the UK AI Security Institute and Meridian Labs. It is used for a broad range of evaluations measuring coding, agentic tasks, reasoning, knowledge, behaviour and multi-modal understanding.

Evaluations are assembled from composable building blocks (datasets, agents, tools and scorers) that are easy to write and reuse, and more than 200 pre-built evaluations can be run on any model. The Inspect View web tool monitors and visualises evaluation runs and logs, and a VS Code extension helps with writing and debugging evaluations.

Inspect is free and open source under the MIT licence.