A comprehensive benchmark evaluation platform for Software Engineering Efficiency across different AI scaffolds and models.

SWE-Effi provides a standardized platform for evaluating and comparing AI-powered software engineering tools across different scaffolds and language models. Our platform aggregates benchmark results and presents them through an interactive web interface.

🌐 Visit the Live Platform
📝 Submit Your Results

SWE-Effi
├── benchmark
│   └── results
│       └── agent-scaffold-stats
│           ├── agentless/
│           │   ├── GPT-4o-mini-2024-07-18/
│           │   │   ├── combined_stats.json
│           │   │   └── summary_stats.json
│           │   └── qwen3-32B/
│           │       ├── combined_stats.json
│           │       └── summary_stats.json
│           ├── agentless-mini/
│           ├── auto-code-rover/
│           ├── openhands/
│           └── swe-agent/
├── scripts/
│   ├── transform-benchmark.py      # data transformation
│   └── update-website.sh           # easy update script
└── website/
    ├── public/
    │   └── data/
    │         └── benchmark/
    │             └── raw/            # benchmark data
    │                 └── summary/    # benchmark data
    └── src/
        └── docs/
            ├── about/
            └── index.tsx

Want to submit your benchmark results? Follow our submission guide →

1. Clone the repository:

   git clone https://github.com/your-org/swe-effi.git
   cd swe-effi

2. Process benchmark data:

   # Process all new benchmark data
   ./scripts/update-website.sh --auto
   
   # Process specific scaffold/model
   ./scripts/update-website.sh agentless gpt-4
   
   # Validate files before processing
   ./scripts/update-website.sh --validate-only

3. Run the website locally:

   cd website
   npm install
   npm run dev

When contributors submit benchmark results via PR:

1. Review the Pull Request for correctness
2. Validate locally (optional):

   git checkout [pr-branch]
   python3 scripts/transform-benchmark.py --validate-only

3. Merge the PR
4. Update the website:

   ./scripts/update-website.sh --auto

update-website.sh options:

• --auto: Process all available data automatically
• --validate-only: Only validate files, don't transform
• --verbose: Show detailed logs
• --help: Show help information

transform-benchmark.py options:

• --scaffold NAME --model NAME: Process specific combination
• --validate-only: Only validate file format
• --auto: Auto process all data with validation
• --verbose: Show detailed logs

• Python 3 for data processing
• Node.js and npm for website

cd website && npm install

Contributor Results → PR Submission → Validation → Processing → Website Integration

1. Results Collection: Contributors submit via GitHub PRs
2. Validation: Automated checks ensure data quality
3. Processing: Scripts transform data for website consumption
4. Integration: Website automatically displays new results

Results must include:

• combined_stats.json
• summary_stats.json

Apache License 2.0