ИИ-агенты
WorkArena
WorkArena: ServiceNow's benchmark for how well web agents handle everyday knowledge-work tasks.
★ 275 звёздNOASSERTION
Смотреть руководство →AgentBench — проект с открытым исходным кодом из категории «ИИ-агенты» от THUDM (THUDM/AgentBench на GitHub). У него 3 765 звёзд на GitHub и 288 форков, он распространяется по лицензии Apache-2.0 и последний раз обновлялся на GitHub 8 февраля 2026.
A rigorous benchmark for LLMs acting as agents.
Pick one concrete task so you can judge whether the project fits.
Clone the repo and check the prerequisites listed in the README.
Install what it needs and configure credentials or model access.
Run a minimal example end to end before wiring it into your stack.
Tests models as actors in environments.
Peer-reviewed benchmark design.
Diverse tasks beyond static QA.
From a leading academic lab.
Runnable evaluation code.
Permissive licensing.
Download the source and move into the project folder.
git clone https://github.com/THUDM/AgentBench.git
cd AgentBenchSet up a virtual environment and install dependencies.
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txtEvaluate a model on a task subset per the docs.
# see README for the task runnersIt is maintained by THUDM at THUDM/AgentBench, with 288 forks.
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.
You need model access to evaluate; the harness itself is free.
AgentBench covers diverse agentic environments; SWE-bench targets real-world software engineering specifically.
Follow the repo docs to run a task subset against your model.
WorkArena: ServiceNow's benchmark for how well web agents handle everyday knowledge-work tasks.
High-performance serving for the LLMs that power your agents.
OSWorld — benchmark multimodal agents on open-ended tasks in real computer environments.
Open-source red-teaming framework that stress-tests LLMs and AI agents.
Chinese-language course on AI basics: GPU architecture, CUDA, LLMs, and agents.
Uber's ADR: discover, observe, benchmark, and detect threats across enterprise AI agents, deployed at Uber.
Другие проекты из категории «ИИ-агенты» с наибольшим числом звёзд в каталоге LibreAI — реалистичный список альтернатив, если AgentBench вам не подходит.
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
Соседние проекты из категории «ИИ-агенты» в каталоге, ближайшие по числу звёзд к AgentBench — продолжайте изучать полку по обе стороны от него.
Build voice-based LLM agents with a modular, open-source toolkit.
An open-source multimodal retrieval engine for agent pipelines.
A portable AI agent runtime for the JVM — write one @Agent, run it on any Java AI stack.
Google's Python SDK for building AI agents on the Antigravity platform and Gemini models.
Research-backed lifelong memory for LLM agents, text and multimodal.
OpenClaw rebuilt in Go — multi-tenant AI agent teams with 5-layer security in a single binary.
Бесплатно и открыто — забирайте исходный код, ставьте звезду репозиторию и начинайте создавать.