Cognition releases SWE-2 coding model, approaching the frontier at lower cost
AI summary
Cognition released the coding model SWE-2, post-trained on the 2.8T-parameter Kimi K3, scoring 50.0% on FrontierCode 1.1 Main—less than one point below Fable 5.1 at 64% lower cost. The company publicly disclosed SWE-2's training methods, including setting linear cost penalties by reasoning effort level within a single RL training run, length-weighted reward baselines, RL rollout serving optimizations and data improvements, so readers can understand how to use reinforcement learning to shift the entire cost-performance frontier.
Key Takeaways
- Cognition released the coding model SWE-2,post-trained on the 2.
- 8T-parameter Kimi K3,scoring 50.
- 0% on FrontierCode 1.
- 1 Main—less than one point below Fable 5.
- 1 at 64% lower cost.
Quick Facts
- Published
- Updated
- Source
- Cognition 模型 / Devin 博客(网页)
- Category
- Model Releases
- AI score
- 70
官方详细公开了 SWE-2 的训练方法与成本惩罚公式,读者可以借此理解如何用 RL 推移整条成本性能前沿。
Related entities
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-2 coding model, approaching the frontier at lower cost. September 10, 2026. https://new.nat6.net/en/news/cmtym40hx0004roske1zgdkmu/