# Cognition — Introducing SWE-2: Pushing the Pareto Frontier

- Company: Cognition (cognition.com)
- Announced: 2026-09-10
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://cognition.com/blog/swe-2
- Record: https://forck.live/items/9935-introducing-swe-2-pushing-the-pareto-frontier
- Subject: Devin / SWE
- Models affected: SWE-2, SWE-1.7, Kimi K3

Cognition introduced SWE-2, a coding model that achieves 50.0% on FrontierCode 1.1 Main, within one point of Fable 5.1 while being 64% cheaper. SWE-2 is post-trained from Kimi K3, a 2.8T-parameter model, and uses a cost-penalized RL algorithm to train all reasoning-effort levels in a single run. The model is available starting today in Devin Desktop and CLI, and is rolling out on Devin Web and Fusion.

## Evidence

Verbatim from https://cognition.com/blog/swe-2:

> Today we’re introducing SWE-2, our most advanced coding model yet.

---

Record: https://forck.live/items/9935-introducing-swe-2-pushing-the-pareto-frontier
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md