Inspect AI: Framework for Large Language Model Evaluations
Open accessAbstract
Inspect is a fully open-source, extensible framework for rigorous evaluation of large language models (LLMs). It enables comprehensive, reproducible assessments across a broad range of task domains—including coding, reasoning, knowledge, agentic tasks, behaviour, and multimodal understanding—supported by extensive tooling, over 100 pre-built benchmarks, and visualisation utilities. Inspect is designed to provide an excellent developer experience while enabling evaluations that can be reproducibly run at scale.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Other open-source or commercial tools such as DeepEval, InspectAI, Phoenix, and GALILEO provide rich analytics, evaluation orchestration, and debugging capabilities.
“Other open-source or commercial tools such as DeepEval [2], InspectAI [3], Phoenix [1], and GALILEO [25] provide rich analytics, evaluation orchestration, and debugging capabilities.”
From Evaluation and Benchmarking of LLM Agents: A Survey (2025) · cites this paper as [3]
- Authors
- UK AI Security Institute
- Corresponding author
- UK AI Security Institute
- Venue
- Zenodo (CERN European Organization for Nuclear Research)
- Publisher
- Zenodo
- Published
- 2024-05-10
- OpenAlex
- W7126180755
- Cited by
- 0
- References
- 0
- Type
- software
- Language
- en
- Open access
- green
- doi.org
- Licence
- opensource.org/licenses/MIT
- Locations
- Zenodo (CERN European Organization for Nuclear Research)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
- Keywords
- large language models, LLM evaluation, benchmarking, reproducibility, reasoning evaluation, knowledge assessment, multimodal understanding, visualization tools
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
