AI
Cognition's budget coding model trades frontier scores for a fifth of the price
Cognition released SWE-2 Flash, a 276-billion-parameter mixture-of-experts coding model with 12 billion active parameters, post-trained from Thinking Machines' Inkling-Small. The company says a new mid-training stage lifted Terminal-Bench 4.0 from 0.9% to 20.0% and FrontierCode 2.0 from 3.2% to 24.0%, and that Flash runs three times faster than SWE-2 at one-fifth the price. It ships inside Devin, while the weights stay behind a gated partner programme.