Cognition introduces SWE-2 Flash coding model
Cognition AI, Inc. →Cognition introduced SWE-2 Flash and made it available in Devin Desktop and CLI. The model builds on Thinking Machines’ Inkling-Small and adds a mid-training stage before reinforcement learning. Cognition said its training changes broadened the model’s coding, reasoning and cybersecurity capabilities. The release positions the model for everyday engineering work such as automations, code review, feature development and documentation updates. Cognition reported a lower price and faster generation than SWE-2, accompanied by its own benchmark comparisons. The company was developing a program to share model weights securely with trusted industry partners and researchers, while explicitly stating that the weights were not yet being released publicly.