Windsurf unexpectedly announced the launch of its first family of proprietary AI models for software development — SWE-1, SWE-1-lite, and SWE-1-mini. The company, already known for its developer tools, emphasized that the new lineup is optimized not only for coding but for the entire software creation process. Particularly intriguing is the fact that this news comes amid rumors of a three-billion-dollar acquisition deal of Windsurf by OpenAI.
The most powerful model, SWE-1, according to Windsurf, shows results on par with Claude 3.5 Sonnet, GPT-4.1, and Gemini 2.5 Pro in internal tests for programmers. However, SWE-1 still lags behind the latest flagship models, such as Claude 3.7 Sonnet, in complex development tasks. SWE-1-lite and SWE-1-mini will be available to all service users, regardless of the plan, while the full-featured SWE-1 will only be available for paid subscriptions. The cost of use has not yet been announced, but Windsurf emphasized that maintaining SWE-1 is cheaper than Claude 3.5 Sonnet.
Previously, Windsurf and other startups like Cursor and Lovable relied on third-party AI models from OpenAI, Anthropic, and Google. The launch of its own model lineup indicates Windsurf’s ambitions to transition from creating applications to developing powerful foundations for them. In a presentation video, the company’s head of research, Nicolas Moy, emphasized that modern models are mostly geared towards coding, but this is not enough for comprehensive software development.
Windsurf notes that SWE-1 was trained on new types of data and using a special methodology that considers unfinished states, long processes, and working with various interfaces — from terminals to IDEs and the web. The company calls SWE-1 the first proof of concept, hinting at the possibility of even more proprietary models in the future.
