Cognition launches SWE-2 Flash, a faster, cheaper coding model

Cognition unveils SWE-2 Flash, a coding model built on Inkling-Small, available in Devin at one-fifth the price of SWE-2 and three times faster.

By GennnAi·November 4, 2026·1 min read

SWE-2 Flash SWE-2 Devin Inkling-Small Thinking Machines Modèles Sécurité Flash

Thanks to a mid-training stage added before reinforcement learning, Terminal-Bench 4.0 rises from 0.9% to 20.0% and FrontierCode 2.0 from 3.2% to 24.0%. The base model, from Thinking Machines, has 276 billion parameters, 12 billion of them active. Cognition is not publishing the weights yet, citing cyber misuse risks, and is building a program to share them securely with partners and researchers. It is available in Devin Desktop and CLI.

Source: Cognition, published on November 4, 2026.