# THUDM/AgentBench

A Comprehensive Benchmark to Evaluate LLMs as Agents (ICLR'24)

Repository: https://github.com/THUDM/AgentBench
Canonical: https://ross.abutalabs.com/products/agentbench
Language: Python
License: Apache-2.0
License Family: permissive
Topics: chatgpt, gpt-4, llm, llm-agent
Last push: 2026-02-08T17:01:05+00:00

## Health v2 (maintenance only)
Score: 58/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 66, release rhythm 35, longevity 80
- inputs: {"age_days": 1132, "days_push": 206, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3695, forks 276 (observed 2026-08-28T04:08:14.871611+00:00)

## What it is
AgentBench is a comprehensive benchmark for evaluating large language models as agents across multi-turn interactive tasks like database querying, OS interaction, web shopping, and knowledge graph reasoning. The current version integrates with AgentRL and uses function-calling style prompts with fully containerized Docker Compose deployment.

## Use cases
- benchmark how well LLMs perform as agents on multi-turn tasks
- compare function-calling abilities of different LLMs
- evaluate agents on OS interaction and database tasks
- run containerized agent evaluation environments
- train agents with RL using AgentRL integration

## When to choose
- you need standardized multi-environment evaluation of LLM agents
- you want reproducible, containerized agent benchmarks
- you are comparing function-calling performance across models

## When to avoid
- you need a simple single-task eval harness
- you cannot run Docker or provide large Freebase data
- you only want static dataset benchmarks without interactive environments

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, agent-framework, llm-inference, testing
- domain: large-language-models, machine-learning, developer-tools
- platform: python, cross-platform
- tags: llm-agents, evaluation, function-calling, reinforcement-learning, leaderboard, iclr, ai-agents, docker, linux

## Member repositories
- THUDM/AgentBench (main) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:14.871611+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:31:05.159501+00:00, confidence not recorded.
  - readme: https://github.com/THUDM/AgentBench (fetched 2026-08-28T04:08:14.871611+00:00, sha ce80fc649420)
- Data as of 2026-08-30T08:39:29.467469+00:00.