Windsurf, a startup known for its AI tools that help software engineers write and edit code through chat, recently launched its own family of AI models called SWE-1. This new line includes SWE-1, SWE-1-lite, and SWE-1-mini, which are designed to support the entire software engineering process, not just coding. This marks a shift for Windsurf from focusing solely on applications to also developing the models behind them.
The introduction of SWE-1 models follows reports that OpenAI is close to acquiring Windsurf for $3 billion, which surprised some in the industry. SWE-1, the most advanced model in the family, performs well on internal coding tests, rivaling models like Claude 3.5 Sonnet, GPT-4.1, and Gemini 2.5 Pro. However, it still falls short of cutting-edge models such as Claude 3.7 Sonnet in certain tasks. SWE-1-lite and SWE-1-mini will be available to all Windsurf users, while the full SWE-1 model will be reserved for paid subscribers.
WindSurf’s approach aims to go beyond traditional coding models by addressing the broader demands of software engineering. Nicholas Moy, Windsurf’s Head of Research, explained that existing AI models excel at writing code but struggle to manage the multiple environments software engineers use, such as terminals, development environments, and online resources. SWE-1 was trained with a new method that incorporates these real-world complexities, like incomplete workflows and multitasking across different tools.
This launch reflects Windsurf’s commitment to differentiating itself from competitors like Cursor and Lovable, which mostly rely on AI models from OpenAI, Anthropic, and Google. By building their own AI models tailored to the entire software engineering process, Windsurf hopes to improve productivity and better meet the needs of developers. They describe SWE-1 as a “proof of concept,” indicating plans to continue developing and releasing new AI models in the future.
As AI continues to transform software development, Windsurf’s move highlights the growing trend of companies creating specialised AI tools that go beyond simple coding assistance. This approach promises to make software engineering more efficient by supporting complex workflows and providing more intelligent assistance, ultimately helping developers build better software faster.
