SWE-1.7

SWE-1.7 is a proprietary AI model developed by Cognition specifically for autonomous software engineering. Unlike general-purpose language models, it is trained to solve complex coding tasks that require planning, repository understanding, debugging, testing, and multi-step execution over extended periods.

Built on a reinforcement learning pipeline and optimized for long-horizon software engineering workflows, SWE-1.7 powers the Devin AI software engineer. The model emphasizes careful codebase exploration, root-cause analysis, and reliable code modifications rather than producing quick code snippets. It supports tool use, persistent task execution, and self-compaction techniques that allow it to work effectively on tasks lasting several hours.

With a 256K token context window, SWE-1.7 is designed for large repositories and complex engineering projects. It is available through Devin across web, desktop, and CLI environments, targeting professional software teams and enterprise development workflows instead of direct consumer chatbot use. According to Cognition, SWE-1.7 achieves frontier-level performance on software engineering benchmarks while significantly reducing inference cost compared with previous generations.