By The Cognition Team09.10.26 — Cognition has introduced SWE-2, their most advanced coding model yet. It achieves 50.0% on Frontier Code 1.1 Main, within one point of Fable 5.1 while being 64% cheaper. Unlike previous models, SWE-2 scales RL to the multi-trillion-parameter regime using an algorithm that trains all reasoning-effort levels in a single run, advancing the entire cost–performance frontier.

On Frontier Code 1.1 Main and Deep SWE 1.1, SWE-2 beats SWE-1.7 and Grok 4.6 on both score and cost, matches GPT-5.6 Sol and Fable 5/5.1 at a fraction of their price, and comes within a few points of GPT-6 Astra at a quarter of the cost. Built on the 2.8T-parameter Kimi K33 model, SWE-2 uses cost penalties tuned to local Pareto frontier slopes, improved reward baselines, and optimized RL rollout serving. It's available today in Devin Desktop and CLI, with rollout on Devin Web and Fusion.

The model demonstrates stronger engineering judgment, writing more complete solutions with fewer detours—scoring higher than SWE-1.7 while taking 58% fewer turns and costing 81% less.