What is GAIA?
GAIA ("General AI Assistants") is a benchmark for evaluating AI agents on real-world tasks. The tasks demand multi-step reasoning, tool use, multimodal handling, and, for many questions, live web browsing or search to find and verify facts. Researchers from Meta AI's FAIR team and Hugging Face built it and published the work as an arXiv paper in November 2023. The benchmark lives entirely on Hugging Face as a 932-row dataset split into three difficulty levels, with a public leaderboard Space where teams submit agent scores.
Our verdict
GAIA is a widely cited reference point for general-purpose agent evaluation, especially for tasks that combine web search with multi-step tool use, and it's regularly grouped with SWE-bench and WebArena in 2026 agent roundups. It's a static research dataset with a community leaderboard rather than a hosted evaluation service, so teams have to run their own agent against the dataset and submit results manually.
Categories:
These are research benchmarks, not products you buy. They define fixed web tasks and score how reliably an agent or model completes them, which is the closest thing the field has to an objective measure of web-agent capability. They are listed here for evaluation, not procurement.
