Skip to content
Back to the index
35Coding agentProvisional score

SWE-agent

SWE-agent

Princeton research project that autonomously resolves GitHub issues using any language model you supply.

Updated

Visit SWE-agent
SWE-agent preview

The read

SWE-agent is an open-source research tool from Princeton and Stanford that gives a language model a structured interface to browse repos, read files, edit code, and run tests in order to fix GitHub issues autonomously. It scores above 74 percent on SWE-bench Verified and has spawned mini-swe-agent, a 100-line successor with comparable benchmark results. Configuration is done through a single YAML file, keeping it flexible for custom tasks including security vulnerability research. Active development is in maintenance mode for the original repo while the team focuses on mini-swe-agent, so teams wanting a fully productionized experience should look elsewhere.

Where it shines

  • Open-source with a strong SWE-bench benchmark pedigree
  • Single YAML config covers GitHub issues, security tasks, and custom workflows
  • Works with any LLM including Claude, GPT, and open models

Worth knowing first

  • Original repo in maintenance-only mode, succeeded by mini-swe-agent
  • No hosted service or UI, requires local setup and API keys

Index score

74.4

Provisional. This score is read from the product's public capability, not from the one-prompt rebuild the fully reviewed entries went through, so treat it as a placing rather than a verdict.

Craft70
Speed62
Control78
Value92
Rank
35 of 54
Builds
Tools
Category
Coding agent
Output
Automated GitHub issue fixes
Pricing
Free tier
Best for
Researchers and engineers evaluating benchmark-level autonomy

Scores are our own editorial judgement, weighted craft 30, speed 25, control 25 and value 20.

Also worth a look

Full index