Windsurf, the startup best known for its AI-powered code assistant tools, has launched its first family of in-house AI models designed specifically for software engineers. The release of the new SWE-1 AI model series — which includes SWE-1, SWE-1-lite, and SWE-1-mini — marks a major leap for the company as it moves from building apps to developing the AI infrastructure that powers them.
The timing of the launch is especially intriguing, coming just as reports surface that OpenAI is acquiring Windsurf in a $3 billion deal. The launch of SWE-1, however, suggests Windsurf is carving out its own path—potentially laying the foundation to become a model developer, not just an app layer.
SWE-1 Aims Beyond Code—Into the Full Software Engineering Lifecycle
What sets SWE-1 apart, according to Windsurf, is that it’s not just optimized for code generation. It’s built to handle the entire software engineering process—a task that requires working across multiple tools, surfaces, and incomplete workflows.
“The best models today are great at writing code,” said Nicholas Moy, Head of Research at Windsurf. “But coding is not software engineering. SWE-1 was designed for the way engineers actually work: switching between IDEs, terminals, web searches, and long-running tasks.”
To train SWE-1, Windsurf used a custom data model and a novel “training recipe” that captures the complexities of real-world development environments—such as incomplete program states, interrupt-driven work, and multi-surface workflows.
Windsurf claims SWE-1 delivers performance on par with top-tier models like Claude 3.5 Sonnet, GPT-4.1, and Gemini 2.5 Pro on internal programming benchmarks. However, it still falls short of frontier-level models like Claude 3.7 Sonnet on some software engineering tasks.
Here’s how the lineup breaks down:
- SWE-1: Full-featured, high-performance model; available only to paid users
- SWE-1-lite & SWE-1-mini: Lighter models accessible to both free and paid users
While Windsurf hasn’t released pricing details yet, it says SWE-1 is cheaper to serve than Claude 3.5 Sonnet—positioning it as a cost-effective alternative for dev-focused AI applications.
From Vibe Coding to Full-Stack AI
Windsurf made its name in the “vibe coding” movement, where engineers collaborate with chat-based AIs to write, refactor, and navigate codebases. Alongside companies like Cursor and Lovable, Windsurf built tools on top of third-party models from OpenAI, Anthropic, and Google.
With the launch of its SWE-1 AI model family, Windsurf is moving into first-party model development—a significant shift in the rapidly growing AI coding space.
This release is described as an “initial proof of concept,” suggesting more models may follow. And if SWE-1 is any indicator, Windsurf is betting that the future of AI in engineering lies beyond just writing lines of code—it’s about helping devs manage the entire lifecycle of modern software projects.
