Cognition Releases SWE-2 Coding Model for Devin
Cognition has introduced SWE-2, its most capable coding model so far and the successor to SWE-1.7. The model was post-trained with reinforcement learning from Kimi K3, Moonshot AI’s 2.8T-parameter open model. Cognition reports a 50.0% result on FrontierCode 1.1 Main, within one percentage point of Fable 5.1 while costing 64% less.
SWE-2 is the company’s first model with selectable reasoning-effort levels. Medium, high and max settings were trained in one reinforcement-learning run, with separate cost penalties. On FrontierCode 1.1 Main, the medium setting used 58% fewer turns and cost 81% less than SWE-1.7, while mean steps fell from 127 to 53.
The model leads its Kimi K3 base on all listed benchmarks and performs best among the compared systems on Terminal-Bench 2.1, but Terminal-Bench 4 remains a weakness. SWE-2 is not available with open weights or as a standalone API. It currently runs in Devin Desktop and CLI, with Devin Web and Fusion being rolled out. Paid-tier access is listed as free through October 10, 2026.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.