← all datasets
◆ Evals & benchmarks
SWE-bench
MITReal GitHub issues with tests, used to measure whether a coding agent can resolve them end to end.
Summary
- Formats
jsonl- License
- MIT
- Added
- 2026-09-10
Get it
From the project’s own instructions where it documents any; otherwise a plain clone.
-
$ git clone https://github.com/SWE-bench/SWE-bench.git
-
$ cd SWE-bench && pip install -e .
Evaluation runs in Docker, so install that first. The task instances are on Hugging Face (SWE-bench, Lite, Verified, Multimodal).
Contents
679 files 78.7 MB repository
-
.yml507 -
.py109 -
.md25 -
.lock17 -
.png4 -
.sh4 -
.svg4 -
.css2
Use it with
Commands are curated, not yet run by us.
jq — Stream the records one per line to see their shape.
$ jq -c . sample.jsonl
DuckDB — Query the file in place with SQL — DuckDB detects CSV, JSON and JSONL by extension.
$ duckdb -c 'SELECT * FROM '\''sample.jsonl'\'' LIMIT 10'