Cognition introduces SWE-2 for Devin

10 Sep 2026: Cognition’s coding model for Devin. The company claims 50.0% on FrontierCode 1.1 Main — within about a point of Fable 5.1 — at 64% less cost. Those are Cognition’s numbers.

10 Sep 2026: Cognition published “Introducing SWE-2,” calling it the company’s most advanced coding model yet. The post says it is available that day in Devin Desktop and CLI, with Devin Web and Fusion rolling out.

Cognition’s headline claim: 50.0% on FrontierCode 1.1 Main, “within one point of Fable 5.1” (50.9% on the same table) and “64% cheaper.” Score and cost cut are Cognition’s.

The same Cognition table: DeepSWE 1.1 at 73.0%, Terminal-Bench 2.1 at 92.8%, Terminal-Bench 4 at 27.3% — versus 55.8% for Fable 5.1 on that last row. Still Cognition’s numbers.

Cognition says SWE-2 is post-trained from Kimi K3, a 2.8-trillion-parameter model that had already had extensive RL for agentic coding. The post calls this the first time the company scaled that RL recipe to the multi-trillion-parameter regime.

No public weights and no standalone per-token API are described on the page. Use is inside Devin. This desk did not rerun FrontierCode.

A dated Devin-model launch. File the availability. Keep the benches and the 64% cheaper line as Cognition’s.