AI Agents
WorkArena
WorkArena: ServiceNow's benchmark for how well web agents handle everyday knowledge-work tasks.
★ 275 starsNOASSERTION
View guide →// agents · THUDM/AgentBench
AgentBench
AgentBench is an open-source ai agents project by THUDM (THUDM/AgentBench on GitHub). It has 3,765 GitHub stars and 288 forks, is released under the Apache-2.0 license, and was last updated on GitHub on 8 February 2026.
A rigorous benchmark for LLMs acting as agents.
★ 3.8k stars⑂ 288 forksApache-2.0updated 2026-02
▦ visual — this project's README ships no embeddable screenshot,
so here's a placeholder. The guide below is complete regardless.
so here's a placeholder. The guide below is complete regardless.
// project facts
Project facts
- Repository
- THUDM/AgentBench
- GitHub stars
- 3,765
- Forks
- 288
- License
- Apache-2.0
- Last updated on GitHub
- 8 February 2026
- Category
- AI Agents
- Listing
- Full install guide
// how it works
How it works
- Start with a use case
Pick one concrete task so you can judge whether the project fits.
- Fetch the source
Clone the repo and check the prerequisites listed in the README.
- Set up the environment
Install what it needs and configure credentials or model access.
- Try the sample workflow
Run a minimal example end to end before wiring it into your stack.
// features
Why people love it
Agentic evaluation
Tests models as actors in environments.
ICLR 2024
Peer-reviewed benchmark design.
Multi-environment
Diverse tasks beyond static QA.
THUDM research
From a leading academic lab.
Python harness
Runnable evaluation code.
Apache-2.0
Permissive licensing.
Advertisement
// run it
Run it yourself
- Step 1 — Clone the repo
Download the source and move into the project folder.
git clone https://github.com/THUDM/AgentBench.git cd AgentBench - Step 2 — Install
Set up a virtual environment and install dependencies.
python -m venv .venv && source .venv/bin/activate pip install -r requirements.txt - Step 3 — Run tasks
Evaluate a model on a task subset per the docs.
# see README for the task runners
// use cases
What you can build
- Evaluate a new model as an autonomous agent.
- Compare agent frameworks on standard tasks.
- Reproduce ICLR 2024 agent results.
- Design better agentic evals for your own models.
Advertisement
// faq
Questions, answered
Who maintains AgentBench?
It is maintained by THUDM at THUDM/AgentBench, with 288 forks.
Is AgentBench free to use?
Yes. It is released under the Apache-2.0 license, which permits commercial use, modification, and redistribution. Read the LICENSE file for the exact terms.
Do I need a paid API to use it?
You need model access to evaluate; the harness itself is free.
How does it compare to SWE-bench?
AgentBench covers diverse agentic environments; SWE-bench targets real-world software engineering specifically.
How do I get started?
Follow the repo docs to run a task subset against your model.
// keep exploring
Similar tools
AI Agents
sglang
High-performance serving for the LLMs that power your agents.
★ 36.9k starsApache-2.0
View guide →AI Agents
OSWorld
OSWorld — benchmark multimodal agents on open-ended tasks in real computer environments.
★ 3.2k starsApache-2.0
View guide →Agents
deepteam
Open-source red-teaming framework that stress-tests LLMs and AI agents.
★ 3.0k starsApache-2.0
View guide →Agents
AI-fundamentals
Chinese-language course on AI basics: GPU architecture, CUDA, LLMs, and agents.
★ 2.9k starsApache-2.0
View guide →AI Agents
ADR
Uber's ADR: discover, observe, benchmark, and detect threats across enterprise AI agents, deployed at Uber.
★ 1.9k starsApache-2.0
View guide →// alternatives
Alternatives in AI Agents
The most-starred other AI Agents projects in the LibreAI catalogue — the realistic shortlist if AgentBench is not the right fit.
- openclaw★ 391.6k
A personal AI that really does things — on any OS, with your data staying yours.
- superpowers★ 294.9k
An agentic skills framework & software development methodology that works.
- ECC★ 275.2k
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
- hermes-agent★ 252.0k
The agent that grows with you — a Python framework for adaptive AI agents.
- deepseek-harness★ 245.5k
DeepSeek's plugin-based agent harness where everything is a plugin.
// more in agents
More in AI Agents
Nearby AI Agents projects in the catalogue, closest in stars to AgentBench — keep exploring the shelf either side of it.
- vocode-core★ 3.8k
Build voice-based LLM agents with a modular, open-source toolkit.
- morphik-core★ 3.7k
An open-source multimodal retrieval engine for agent pipelines.
- atmosphere★ 3.8k
A portable AI agent runtime for the JVM — write one @Agent, run it on any Java AI stack.
- antigravity-sdk-python★ 3.7k
Google's Python SDK for building AI agents on the Antigravity platform and Gemini models.
- SimpleMem★ 3.8k
Research-backed lifelong memory for LLM agents, text and multimodal.
- goclaw★ 3.6k
OpenClaw rebuilt in Go — multi-tenant AI agent teams with 5-layer security in a single binary.
Get AgentBench
Free and open — grab the source, star the repo, and start creating.