SWE-2 Flash

SWE-2 Flash is Cognition's fast, low-cost coding model, built on Thinking Machines' Inkling-Small (276B total / 12B active parameters) with a large-scale mid-training stage (a few trillion tokens of scientific articles, blogs, code and other sources, in three stages) followed by large-scale reinforcement learning on higher-quality, more diverse task types. Cognition reports 24.0% on FrontierCode 2.0, 45.5% on FrontierCode 1.1 Main, 60.1% on DeepSWE 1.1, 20.0% on Terminal-Bench 4.0 and 89.2% on Terminal-Bench 2.1, and states it outperforms all American open-weight models. Training targeted cybersecurity work (ExploitBench 34.8%, up from 11.3% for the base model). It is available starting 2026-11-04 in Devin Desktop and Devin CLI (free through 2026-11-30) at one-fifth the price and about 3x the generation speed (roughly 200 vs 60 tokens/s) of SWE-2. Weights are not public; Cognition plans a program to share them with trusted industry partners and researchers. No standalone public API, per-token pricing, or context-window specification has been published.