What is DeepSWE?
Datacurve's open benchmark and leaderboard ranking coding agents on original long-horizon tasks, with cost per task.
Who is DeepSWE best for?
DeepSWE is best for engineering leads picking a coding model who want scores on unseen tasks weighed against cost.
Picking a coding model
The leaderboard plots score against average cost per task, so you can see which model is the cheapest at a given level.
Avoiding contaminated scores
Tasks are written from scratch rather than taken from public pull requests, so models cannot have seen the answers in training.
Testing your own agent
The open tasks run with Datacurve's Pier tool on mini-swe-agent, Claude Code, Codex, Gemini CLI or opencode.
Agent researchers
Each run records output tokens and agent steps, and the write-up covers how leading models fail.
How much does DeepSWE cost?
DeepSWE is free: the leaderboard is public and the tasks and code are open source under Apache 2.0, though running it yourself costs whatever your model API calls cost.
- Public leaderboard
- 113 tasks across 91 repositories
- Runs with your own model API keys
There is no paid tier. Model providers bill your own API usage when you run the tasks, and leaderboard submissions go to the Datacurve team by email. Full pricing on deepswe.datacurve.ai. We checked these prices on .
How do readers rate DeepSWE?
No one has rated DeepSWE yet. Be the first.
Sign in to rate. One rating per account.
Is DeepSWE open source?
Yes. DeepSWE's code is public under the Apache 2.0 licence, so you can read it, change it and run it yourself.
On GitHub
1,924stars
- Last commit
- 26 Aug
- Repository age
- 6 months
- Licence
- Apache 2.0
- Self-hosted
- No
- Repository
- datacurve-ai/deep-swe
Questions about DeepSWE
What is DeepSWE?
Datacurve's open benchmark and leaderboard ranking coding agents on original long-horizon tasks, with cost per task.
Who is DeepSWE for?
DeepSWE is best for engineering leads picking a coding model who want scores on unseen tasks weighed against cost.
Is DeepSWE free?
Yes. DeepSWE has a free Open source plan: public leaderboard.
Is DeepSWE open source?
Yes. DeepSWE's code is public under the Apache 2.0 licence, so you can read it, change it and run it yourself.
Where can I try DeepSWE?
On its official website, deepswe.datacurve.ai.
How is DeepSWE different from SWE-bench Pro?
DeepSWE's tasks are written from scratch instead of adapted from existing commits, and they use hand-written checks of software behaviour. Datacurve says its prompts are half the length of SWE-bench Pro's while solutions need about 5.5 times more code.







