OpenSWE released 45,320 executable SWE training environments, and the difficulty filter is the part worth copying
AspireAI community · arXiv (Fu et al. - daVinci-Env)
Training capable software engineering agents demands large-scale, executable and verifiable environments that provide dynamic feedback loops for iterative code editing, test execution and solution refinement. Open-source datasets remain limited in scale and repository diversity while industrial solutions are opaque, so OpenSWE presents the largest fully transparent framework for SWE agent training in Python: 45,320 executable Docker environments spanning more than 12,800 repositories, with all Dockerfiles, evaluation scripts and infrastructure open-sourced for reproducibility.
It is built through a multi-agent synthesis pipeline deployed across a 64-node distributed cluster, automating repository exploration, Dockerfile construction, evaluation script generation and iterative test analysis. Beyond scale, a quality-centric filtering pipeline characterizes the inherent difficulty of each environment, filtering out instances that are either unsolvable or insufficiently challenging and retaining only those that maximize learning efficiency. With 891K spent on environment construction and an additional 576K on trajectory sampling and difficulty-aware curation, the project represents a total investment of approximately $1.47 million, yielding about 13,000 curated trajectories from roughly 9,000 quality-guaranteed environments.
OpenSWE-32B and OpenSWE-72B achieve 62.4 percent and 66.0 percent on SWE-bench Verified, establishing state of the art among the Qwen2.5 series. SWE-focused training also yields out-of-domain gains, including up to 12 points on mathematical reasoning and 5 points on science benchmarks, without degrading factual recall.
Caveat: the models are Qwen2.5-based, and the recipe is public while the compute was not cheap.
https://arxiv.org/abs/2603.13023
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.