Arctic-SWE: Revisiting the Long Horizon Agent Training by Modeling-System Co-Design

Key points

• Long-horizon agent training, including software engineering, is advancing rapidly, but the recipe and infrastructure needed for a strong reinforcement-learning baseline remain difficult to reproduce.1
• This testbed exposes failures that early reward gains conceal and guides the integration of established components: gated reward, importance-weight control, aligned training and inference engines, and consistent state across policy updates.1
• On raw R2E-Gym data, without new algorithms or data work, Arctic-SWE trains Qwen3.5-4B to 62.0% pass@1 on SWE-bench Verified.1
• Arctic-SWE provides a documented starting point for studying better algorithms, data and environments for asynchronous agent RL.1

Sources (1)

• [1]Arctic-SWE: Revisiting the Long Horizon Agent Training by Modeling-System Co-Design
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 11:24 PM

  We present Arctic-SWE, a fully asynchronous RL framework that establishes such a baseline through model-system co-design, using overfitting on a small task pool as the testbed.

  Long-horizon agent training, including software engineering, is advancing rapidly, but the recipe and infrastructure needed for a strong reinforcement-learning baseline remain difficult to reproduce.

Extractive summary: sentences quoted from the sources.