Open-SWE-Traces
NVIDIA dataset of software-engineering agent trajectories (OpenHands, SWE-agent, mini-swe-agent, Claude Code, OpenCode) for distilling SWE agents; CC BY 4.0.
Tags
Description
Full message histories, tool definitions, model and gold patches and success metadata for issue-resolution tasks drawn from permissively licensed task sets (Scale-SWE, SWE-rebench-V2). Versioned configs: v1.0 (151k cleaned trajectories from MiniMax and Qwen models; git-hacking runs removed), v1.1 (DeepSeek-V4-Flash, Qwen3.6-27B), v1.2 (Qwen3.8-27B) and v2.0, released 2026-10-26 (Qwen3.8-Flash-Next, xhigh reasoning) with splits per harness. Size category 100K–1M rows; parquet. The card states it is ready for commercial and non-commercial use.
Agent access
Related
- SWE-rebench: Nebius's continuously refreshed, decontaminated SWE agent benchmark and dataset: 21,000+ validated GitHub issue-PR tasks plus a time-windowed leaderboard.
- SWE-smith: Toolkit and datasets for scaling training data for software-engineering agents by synthesizing task instances from GitHub repos (NeurIPS 2025 D&B).
- SWE-Gym: Environment and dataset for training software-engineering agents and verifiers on real GitHub tasks (ICML 2025).
- OpenHands: Open-source, model-agnostic platform for cloud coding agents.
- mini-swe-agent: Minimal open-source software-engineering agent from the SWE-bench/SWE-agent team: about 100 lines of Python, bash as its only tool, a linear history, any model via LiteLLM.
Guides & comparisons
- Agent environment frameworks compared: Harbor vs AgentEnv Framework vs Seahaven (Comparison): Three open-source Python frameworks for building the environments agent evals and RL runs happen in: Harbor (container tasks), Scale AI's AgentEnv Framework and Kiln AI's Seahaven (synthetic SQLite worlds). Covers tasks, grading, compute, agent interface, interop and license.
Sources
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)