From the source
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.






