On May 15, 2025, Windsurf, a leading innovator in AI-driven software development tools, announced the launch of its proprietary family of AI models, collectively named SWE-1. This suite includes SWE-1, SWE-1-lite, and SWE-1-mini, each designed to enhance various facets of the software engineering process beyond mere code generation.
The introduction of these in-house models signifies a strategic shift for Windsurf, moving from solely developing applications to creating the foundational AI models that power them. This development comes amidst reports of OpenAI’s potential $3 billion acquisition of Windsurf, highlighting the startup’s growing influence in the AI coding landscape.
SWE-1, the flagship model, is engineered to perform competitively with existing AI models like Claude 3.5 Sonnet, GPT-4.1, and Gemini 2.5 Pro on internal programming benchmarks. However, it currently falls short of surpassing frontier models such as Claude 3.7 Sonnet in software engineering tasks. Despite this, Windsurf emphasizes that SWE-1 is optimized for the entire software engineering process, addressing tasks that extend beyond code writing to include project management, debugging, and system design.
The SWE-1-lite and SWE-1-mini models are accessible to all users on the Windsurf platform, regardless of their subscription status. In contrast, access to the full capabilities of SWE-1 is reserved for paid subscribers. While specific pricing details have not been disclosed, Windsurf asserts that serving SWE-1 is more cost-effective than utilizing models like Claude 3.5 Sonnet.
Windsurf has gained recognition for its tools that enable software engineers to engage in vibe coding, a method where developers interact with an AI chatbot to write and edit code through conversational prompts. This approach has been embraced by other startups in the space, including Cursor and Lovable. Traditionally, these companies have relied on AI models from providers like OpenAI, Anthropic, and Google.
Nicholas Moy, Windsurf’s Head of Research, highlighted the company’s commitment to advancing beyond current AI capabilities. In a recent announcement, Moy stated, Today’s frontier models are optimized for coding, and they’ve made massive strides over the last couple of years. But they’re not enough for us… Coding is not software engineering. This sentiment underscores Windsurf’s dedication to developing AI solutions that comprehensively address the multifaceted nature of software engineering.
The company notes that while existing models excel at code generation, they often struggle with tasks that require integration across multiple platforms, such as terminals, Integrated Development Environments (IDEs), and web interfaces. To address this, SWE-1 was trained using a novel data model and a training recipe that encapsulates incomplete states, long-running tasks, and multiple surfaces, aiming to provide a more holistic AI assistant for developers.
Windsurf describes SWE-1 as an initial proof of concept, indicating plans for future AI model releases that will further enhance the software engineering process. This initiative reflects the company’s ongoing commitment to innovation and its vision of transforming software development through advanced AI integration.
