<div align="center"> <a href="https://mini-swe-agent.com/latest/"><img src="https://github.com/SWE-agent/mini-swe-agent/raw/main/docs/assets/mini-swe-agent-banner.svg" alt="mini-swe-agent banner" style="height: 7em"/></a> </div>

The minimal AI software engineering agent

📣 New tutorial on building minimal AI agents<br/> 📣 Gemini 3 Pro reaches 74% on SWE-bench verified with mini-swe-agent!<br/> 📣 New blogpost: Randomly switching between GPT-5 and Sonnet 4 boosts performance

[!WARNING] This is mini-swe-agent v2. For the previous version, check out the v1 branch.

In 2024, we built SWE-bench & SWE-agent and helped kickstart the coding agent revolution.

We now ask: What if our agent was 100x smaller, and still worked nearly as well?

The mini agent is for

• Researchers who want to benchmark, fine-tune or RL without assumptions, bloat, or surprises
• Developers who like to own, understand, and modify their tools
• Engineers who want something trivial to sandbox & to deploy anywhere

Here's some details:

• Minimal: Just some 100 lines of python for the agent class (and a bit more for the environment, model, and run script) — no fancy dependencies!
• Performant: Scores >74% on the SWE-bench verified benchmark benchmark; starts much faster than Claude Code
• Deployable: In addition to local envs, you can use docker, podman, singularity, apptainer, and more
• Built by the Princeton & Stanford team behind SWE-bench, SWE-agent, and more (see below)
• Widely adopted: In use by Meta, NVIDIA, Essential AI, Anyscale, and others
• Tested: [![Codecov](https://img.shields.io/codecov/c/git...