Vibe coding startup Windsurf launches in-house AI models

On Thursday, Windsurf, a start-up company that develops the famous artificial intelligence tools for software engineers, announced the launch of the first family of software engineering intelligence, or SWE-1 for a short period. The startup says she trained her new family from Swe-1, Swe-1-Lite and Swe-1-MINI-to improve the “entire software engineering process”, not just coding.

The launch of artificial intelligence models in Windsurf at home may be a shock to some, given that Openai has closed a $ 3 billion deal to acquire Windsurf. However, this model launch indicates that Windsurf is trying to expand beyond the development of applications to develop the models they are also working on.

According to Windsurf, SWE-1, which is the largest and most capable artificial intelligence model, works competitive with Claude 3.5 Sonnet, GPT-4.1 and Gemini 2.5 Pro on internal programming criteria. However, SWE-1 seems less than Frontier AI models, such as Claude 3.7 Sonnet, in software engineering tasks.

Windsurf says Swe-1-Lite and Swe-1-MINI will be available to all users on its platform, free or paid. Meanwhile, Swe-1 will only be available to paid users. Windsurf did not immediately announce the pricing of SWE-1 models, but it claims that it is cheaper to serve Claude 3.5 Sonnet.

Windsurf is famous for the tools that allow software engineers to write and edit software through talks with Chatbot, Amnesty International, a practice known as “atmosphere coding”. Other famous startups of Vibe Corsor include the largest in space, as well as the beloved. Most of these startups, including Windsurf, have traditionally relied on artificial intelligence models from Openai, HotHROPIC and Google to run their applications.

In a video announcement of SWE models, confirming the comments made by the head of Windsurf Research, Nicholas Moy, the latest efforts to distinguish her approach. “The border models have been improved today for coding, and they have taken tremendous steps over the past two years, but they are not enough for us […] Coding is not software engineering. “

Windsurf notes in a blog post that although other models are good in writing code, they are struggling to work between multiple surfaces – as programmers often do – such as stations, IDES, and the Internet. The startup says that SWE-1 has been trained using a new data model and “a recipe for training that envelops incomplete cases, long-term tasks and multiple surfaces.”

Swe-1 starts as a “initial guide to the concept”, indicating that it may launch more artificial intelligence models in the future.