Windsurf, the fast-growing startup behind AI-powered coding tools, has just launched its own family of AI models. The lineup includes SWE-1, SWE-1-lite, and SWE-1-mini — all designed to support the full software engineering workflow, not just coding.
The move marks a bold shift for Windsurf. Until now, the company relied on third-party models like GPT-4 and Claude. But with this launch, Windsurf is stepping into the AI infrastructure game, building the models behind its vibe-coding tools.
SWE-1 is the most advanced of the three. According to the company, it performs on par with models like GPT-4.1, Claude 3.5 Sonnet, and Gemini 2.5 Pro in internal tests. While it doesn’t yet match the newest frontier models, such as Claude 3.7 Sonnet, it still shows strong performance across engineering tasks.
What sets Windsurf apart is its focus. The startup isn’t just optimizing for code generation. It’s targeting the messy, real-world workflow of software engineers — juggling terminals, browsers, and IDEs all at once. SWE-1 was trained with this in mind.
In a launch video, Nicholas Moy, Windsurf’s Head of Research, said coding alone isn’t enough. “Frontier models are great at writing code,” he explained, “but software engineering is more than that.” He described SWE-1 as a model built to handle incomplete tasks, long feedback loops, and context switching across different tools.
Windsurf says SWE-1-lite and SWE-1-mini will be available to both free and paid users. The flagship SWE-1 model will be exclusive to premium subscribers. While exact pricing hasn’t been revealed, the company claims SWE-1 is cheaper to run than Claude 3.5 Sonnet — which could help attract cost-conscious teams.
This launch comes amid reports that OpenAI is finalizing a $3 billion acquisition of Windsurf. That deal hasn’t been confirmed. Still, Windsurf’s decision to develop its own models suggests it’s planning for a more independent future.
The startup is part of a fast-growing group of “vibe coding” tools. These platforms let engineers collaborate with AI in natural language. Other players in the space include Cursor, currently the biggest, and Lovable. Most have built on models from OpenAI or Anthropic. Windsurf’s pivot to in-house AI changes that dynamic.
SWE-1 is described as a “proof of concept,” meaning more models may be on the way. But even this first release shows where Windsurf is headed — toward building full-stack AI that understands how engineers actually work.
