• Windsurf launches SWE-1, SWE-1-lite, and SWE-1-mini AI models for comprehensive software engineering tasks.
  • SWE-1 is available to paid users, while lite and mini versions are open to all.
  • The move marks Windsurf’s expansion from app developer to AI model creator amid reports of an OpenAI acquisition.

Windsurf, known for its conversational “vibe coding” tools for software engineers, has announced the launch of its own AI model suite: SWE-1, SWE-1-lite, and SWE-1-mini.

These models are built to support the full software engineering workflow, from writing code to managing tasks across terminals, IDEs, and the web.

The timing is notable, as reports suggest OpenAI has just closed a $3 billion deal to acquire Windsurf.

The new models signal Windsurf’s ambition to move beyond app development and into building the core AI powering its platform.

According to Windsurf, its flagship SWE-1 model performs competitively with leading models like Claude 3.5 Sonnet, GPT-4.1, and Gemini 2.5 Pro on internal programming benchmarks, though it trails the latest frontier models such as Claude 3.7 Sonnet on some tasks.

SWE-1-lite and SWE-1-mini will be available to all users, while SWE-1 is reserved for paid subscribers. Windsurf claims its models are more cost-effective to serve than some competitors.

The company highlights that while most AI models excel at code generation, SWE-1 is trained to handle the broader, more complex realities of software engineering, including incomplete states and long-running tasks.

Windsurf describes SWE-1 as an “initial proof of concept,” hinting at future model releases to further enhance software engineering productivity.

Edited by Annette George