Cognition releases SWE-grep and SWE-grep-mini, RL-trained fast parallel code retrieval models
Original title:Introducing SWE-grep and SWE-grep-mini: RL for Multi-Turn, Fast Context Retrieval
AI summary
Cognition released SWE-grep and SWE-grep-mini, models focused on highly parallel code context retrieval that match frontier coding models in retrieval capability at an order of magnitude lower latency. The models make up to 8 parallel tool calls per round over a maximum of 4 rounds, are trained with multi-turn reinforcement learning using weighted F1 of file and line-range retrieval as reward; SWE-grep-mini is distilled from SWE-grep and further RL-trained.
Key Takeaways
- Cognition released SWE-grep and SWE-grep-mini,models focused on highly parallel code context…
- The models make up to 8 parallel tool calls per round over a maximum of 4 rounds,are trained with…
- SWE-grep-mini is distilled from SWE-grep and further RL-trained.
Quick Facts
- Published
- Updated
- Source
- Cognition 模型 / Devin 博客(网页)
- Category
- Model Releases
- AI score
- 62
原文给出检索质量分数与推理速度对比数字,并说明多轮 RL 训练方法,读者可评估这种专用检索模型能否迁移到自己的编码工作流。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Cognition releases SWE-grep and SWE-grep-mini, RL-trained fast parallel code retrieval models. October 16, 2025. https://new.nat6.net/en/news/cmtyo2iho03dirog0278hdk11/