Windsurf, a startup known for developing popular AI tools for software engineers, has introduced its first family of proprietary AI software engineering models — the SWE-1 series. The startup states that the SWE-1, SWE-1-lite, and SWE-1-mini models are optimized not only for coding but for the entire software engineering process.
According to Windsurf, the SWE-1 model performs competitively in internal programming benchmarks against Claude 3.5 Sonnet, GPT-4.1, and Gemini 2.5 Pro. However, SWE-1 falls slightly behind frontier AI models, such as Claude 3.7 Sonnet, on software engineering tasks.
The SWE-1-lite and SWE-1-mini models will be available to all users on the platform, both free and paid, while SWE-1 will be accessible exclusively to paid users. Windsurf has not yet disclosed pricing for the SWE-1 models but emphasizes that serving the models is more cost-effective compared to Claude 3.5 Sonnet.
Windsurf is recognized for its vibe-coding tools, which enable software engineers to write and edit code through conversational AI. Other vibe-coding startups include Cursor and Lovable. Traditionally, these companies have relied on AI models from OpenAI, Anthropic, and Google to power their applications.
Nicholas Moy, Head of Research at Windsurf, noted in a presentation video: “Today’s frontier models are optimized for coding, and they have made significant progress in recent years. But that’s not enough for us... Coding is not software engineering.”
According to Windsurf’s blog post, while other models excel at writing code, they struggle to operate across multiple surfaces programmers typically use — such as terminals, IDEs, and the internet. The SWE-1 model was trained using a new training methodology that accounts for incomplete states, long-running tasks, and multiple work environments.
The startup describes SWE-1 as an “initial proof of concept” and plans to develop more AI models in the future.
