Vibe-coding startup Windsurf launches in-house AI models
On Thursday Windsurf, a startup that develops popular AI tools for software, announced The launch of his first family of AI Software Engineering models, or simply SWE-1. The startup says that it has trained its new AI models family to be optimized for the ‘entire software engineering process’, not just coding.
The launch of the internal AI models of Windsurf can come a shock for some, since OpenAi is reportedly Dealed a $ 3 billion deal to acquire windsurf. However, this modeling suggests that Windsurf tries to expand than just developing applications to also develop the models that they feed from electricity.
According to WindsurfSWE-1, the largest and most capable AI model of the couple, performs competitive with Claude 3.5 Sonnet, GPT-4.1 and Gemini 2.5 Pro on internal programming benchmarks. SWE-1, however, seems to be short of frontier AI models, such as Claude 3.7 Sonnet, on software engineering tasks.
Windsurf says that his SWE-1-Lite and SWE-1-Mini models will be available to all users on his platform, free or paid. In the meantime, SWE-1 is only available for paid users. Windsurf did not immediately announce the prices for its SWE-1 models, but claims that it is cheaper to serve than Claude 3.5 Sonnet.
Windsurf is best known for tools with which Software -Tendermeurs can write and edit code through conversations with an AI chatbot, a practice known as ‘atmospheric coding’. Other popular atmospheric startups are cursor, the largest in space, as well as sweet. Most of these startups, including windsurf, traditionally familiarized on AI models from OpenAi, Anthropic and Google to power their applications.
In a video that announces the SWE models, comments from Windsurf’s Head of Research, Nicholas Moy, underline comments, the latest efforts of Windsurf to distinguish his approach. “Today’s Frontier models have been optimized for coding, and they have made enormous progress in recent years,” says Moy. “But they are not enough for us … Coding is not a software engineering.”
Windsurf records in a blog post that although other models are good at writing code, they have difficulty working between multiple surfaces – such as programmers often do – such as terminals, Ides and Internet. The startup says that SWE-1 has been trained with the help of a new data model and a “training recipe that incomes incomplete situations, long-term tasks and multiple surfaces.”
The startup describes SWE-1 as its ‘first proof of concept’, which suggests that it can release more AI models in the future.
