On May 15th, Windsurf announced the release of Wave9 update, officially launching its first series of AI models, SWE-1, specifically optimized for the entire software engineering process. The series includes three models: SWE-1, SWE-1-lite, and SWE-1-mini, aiming to enhance software development efficiency and claims to accelerate up to 99%.
According to internal evaluations by Windsurf, the performance of the SWE-1 series is close to the leading Frontier model in the industry. It performs comparably with models like Claude3.5Sonnet and GPT-4.1 on software engineering tasks, particularly excelling in handling terminal operations, testing, and user feedback in complex scenarios. The main SWE-1 model supports advanced reasoning and tool usage, accessible without limits only to paying users; SWE-1-lite replaces the original Cascade Base and is freely available to all users; SWE-1-mini provides quick code predictions for Windsurf Tab and is also unrestricted for all users.
The core innovation of the SWE-1 series lies in "flow awareness," enabling seamless collaboration between AI and developers on a shared timeline, covering multiple interfaces, unfinished tasks, and long-term projects. Windsurf emphasizes that coding is just part of software engineering. The SWE-1 series, through customized training, breaks through the limitations of existing AI models and optimizes the entire development process.
Currently, the SWE-1 series has been integrated into the Windsurf editor, and developers can immediately experience it at windsurf.com. Notably, there are rumors in the market that OpenAI will acquire Windsurf for $3 billion, but the deal has not yet been officially confirmed. The launch of SWE-1 marks Windsurf's strategic transformation from a tool provider to a bottom-level model developer. In the future, the company plans to further expand its machine learning research team.