AgentBench

Editor note

Tests an LLM acting as an agent across eight interactive environments in the original paper, such as an operating system shell, databases, a knowledge graph and web shopping. The current repo is a function-calling version that runs five of these tasks in Docker; older releases stay available as v0.1 and v0.2.

What it tests

Multi-turn agent reasoning and decision-making across varied environments.

Original source

github.com/THUDM/AgentBench

Last updated

Repo last push 2026-02-08

License

Apache-2.0

Facts come only from the linked source (checked 15–16 Oct 2026). Credit belongs to the creators. We don't score or rank evals: how we review.

Reviews

No reviews yet.

Reviews are written by people who sign in with GitHub, one per account per eval. We can't confirm that each reviewer used the eval; we remove reviews that break our rules, never just for being negative, and we never add, buy or edit reviews. Averages, when shown, come only from these reviews.

Related agents & tool use evals

• GAIA — General assistant skills: browsing, file handling, multi-step reasoning.
• WebArena — Autonomous web navigation and task completion.
• OSWorld — Computer-use agents operating real operating systems and apps.
• Terminal-Bench — Agents working in the shell end to end.

All Agents & tool use evals · Comparison table