Windsurf Launches In-House AI Models for Software Engineering

Windsurf Unveils New AI Models for Enhanced Software Engineering

Windsurf, a startup known for its AI-powered tools for software engineers, has launched its first family of in-house AI models: SWE-1, SWE-1-lite, and SWE-1-mini. These models are designed to optimize the entire software engineering workflow, going beyond simple code generation.

This launch is notable given OpenAI's reported $3 billion deal to acquire Windsurf. It suggests Windsurf is expanding its focus from application development to include creating the underlying AI models.

SWE-1: Performance and Availability

According to Windsurf, SWE-1, the most powerful model in the family, rivals Claude 3.5 Sonnet, GPT-4.1, and Gemini 2.5 Pro on internal programming benchmarks. While it performs competitively with these models, it currently trails frontier models like Claude 3.7 Sonnet on specific software engineering tasks.

SWE-1-lite and SWE-1-mini will be available to all Windsurf users, both free and paid. SWE-1 will be exclusive to paid users. Pricing details for SWE-1 are yet to be released, but Windsurf claims it's more cost-effective than Claude 3.5 Sonnet.

Beyond Vibe Coding: A New Approach

Windsurf has been a key player in "vibe coding," allowing engineers to write and edit code conversationally with an AI chatbot. Like other vibe-coding startups such as Cursor and Lovable, Windsurf has traditionally relied on models from OpenAI, Anthropic, and Google.

"Today's frontier models are optimized for coding, and they've made massive strides over the last couple of years," says Nicholas Moy, Windsurf's Head of Research. "But they're not enough for us — coding is not software engineering."

Windsurf emphasizes that while existing models excel at writing code, they often struggle with the multifaceted nature of software engineering, which involves working across terminals, IDEs, and the internet. SWE-1 was trained with a new data model and training process designed to address these complexities, handling incomplete states, long-running tasks, and multiple work surfaces.

Windsurf considers SWE-1 an "initial proof of concept," indicating the potential for future AI model releases.