Evals

AgentBench

AgentBench is the first benchmark designed to evaluate LLM-as-Agent across a diverse spectrum of environments. It bundles 8 environments — 5 newly created (Operating System, Database, Knowledge Graph, Digital Card Game, Lateral Thinking Puzzles) and 3 adapted (House-Holding from ALFWorld, Web Shopping from WebShop, Web Browsing from Mind2Web). The benchmark requires roughly 4,000 dev-set and 13,000 test-set interactions per model.

GitHub

SDKs

📦
GitHubRepository
https://github.com/THUDM/AgentBench

Other Resources

🔗
Paper
https://arxiv.org/abs/2308.03688
🔗
Leaderboard
https://llmbench.ai/agent

Work with this as data

Every API here is available over the APIs.io API and to AI agents over MCP.

MCP server

One button, every client — Claude, Cursor, VS Code and the rest.

https://apis.io/mcp

Tools for apis

7 MCP tools reach this
  • find_apisBrowse and filter every API in the catalog.
  • get_api_artifactsOne API's artifacts, grouped by type.
  • get_openapiThe primary OpenAPI for this API.
  • find_similar_apisAPIs that look like this one.
  • apis_io_searchSTART HERE — APIs, providers and tags for one query, each with its total.
  • resolveTurn a domain, URL or GitHub org into the provider it belongs to.
  • find_cohortsEvery scored population of providers in the catalog.
All 92 tools →

Call it yourself

curl for this page
This API 
curl "https://apis.io/api/v1/apis/evals:agentbench"
All apis 
curl "https://apis.io/api/v1/apis?limit=25"

Calls without X-API-Key need no key. Calls with it need a paid plan — get a key.

Get an API key

Free tier, no form to fill in. Signing in shares your email address with us — we store it to create your key and to recognise you if you sign in with another provider. See our Privacy Policy and Terms.

GitHub Google LinkedIn

A second provider on the same verified email joins the account you already have.

API entry from apis.yml

apis.yml Raw ↑ 
name: AgentBench
description: AgentBench is the first benchmark designed to evaluate LLM-as-Agent across a diverse spectrum
  of environments. It bundles 8 environments — 5 newly created (Operating System, Database, Knowledge
  Graph, Digital Card Game, Lateral Thinking Puzzles) and 3 adapted (House-Holding from ALFWorld, Web
  Shopping from WebShop, Web Browsing from Mind2Web). The benchmark requires roughly 4,000 dev-set and
  13,000 test-set interactions per model.
humanURL: https://github.com/THUDM/AgentBench
baseURL: https://github.com/THUDM/AgentBench
tags:
- Benchmarks
- AI Agents
- Multi-Environment
- LLM-as-Agent
- Tsinghua
tags_raw:
- Benchmark
- AI Agents
- Multi-Environment
- LLM-as-Agent
- Tsinghua
properties:
- type: GitHubRepository
  url: https://github.com/THUDM/AgentBench
- type: Paper
  url: https://arxiv.org/abs/2308.03688
- type: Leaderboard
  url: https://llmbench.ai/agent