Mini toolsSWE 1.7
By Cognition AI
SWE 1.7 is an agentic coding model built by Cognition through extensive reinforcement learning on top of a Kimi K2.7 base. It is optimized for long horizon asynchronous software engineering work, using self compaction to summarize its working state and continue tasks well beyond its raw context window, with training rollouts reaching up to six hours. The model explores codebases more thoroughly than prior models, making more tool calls, file reads, and searches before acting, and investigates root causes, edge cases, and hidden requirements rather than guessing. It produces condensed, efficient chain of thought reasoning as a result of an alternating length penalty applied during training. SWE 1.7 is available today inside Devin (Web, Desktop, and CLI), served via Cerebras at 1000 tokens per second.
New
Coding
Overview
SWE 1.7 is Cognition's most capable coding model to date, reaching frontier level intelligence at a much lower cost. Trained with reinforcement learning from a Kimi K2.7 base, it is optimized for long horizon, asynchronous agentic software engineering tasks and thorough codebase exploration.
About Cognition AI
Cognition AI is an applied artificial intelligence lab focused on reasoning and coding, best known for creating Devin, recognized as the world's first autonomous AI software engineer. Founded in 2023 by Scott Wu, Walden Yan, and Steven Hao, the company focuses on developing AI agents that act as collaborative team members capable of planning and executing complex software engineering tasks
Industry:
Artificial Intelligence
Company Size:
168
Location:
San Francisco, CA, US
Website:
cognition.ai
Last updated: July 8, 2026
