Windsurf Unveils SWE-1 AI Model Family to Transform End-to-End Software Engineering

Vibe-coding platform Windsurf announced on Thursday the debut of its first family of AI software-engineering models, known as SWE-1.

The company claims that it has trained its new family of AI models SWE-1, SWE-1-lite and SWE-1-mini to be optimised for the “entire software engineering process,” not just coding.

“Today, we are launching our first family of models, dubbed SWE-1, optimised for the entire software engineering process, not just the task of coding,” read the startup’s blog.

The release of Windsurf’s in-house AI models comes just days after OpenAI reportedly finalised a US$3 bn acquisition deal for Windsurf. However, this model introduction suggests that Windsurf is aiming to expand beyond merely developing programmes to include the models that power them.

The company describes the SWE-1 model as one that matches Claude 3.5 Sonnet’s tool-call logic at a lower serving cost and is available free of charge to all paying users during its promotional period. The SWE-1-lite is a more compact model that surpasses Cascade Base and is accessible to all users, free or paid, without restrictions, while the SWE-1-mini is an ultra-fast, lightweight model that supports the Windsurf Tab’s passive experience for all users regardless of subscription status.

Software Development Suite

Windsurf stated that while code-capable models have made significant advancements over time, progressing from simple code writing to one-shot application generation, they will soon reach a saturation point.

As per the startup, real software engineering involves far more than writing code, it requires working in the terminal, sourcing external information, iterating on products, testing and interpreting user feedback. Current models excel at tactical short-horizon tasks, generating code that compiles or passes unit tests but struggle with the long-term multi-stage reasoning needed to maintain and evolve complex systems.

Without the ability to navigate incomplete project states and handle ambiguity over extended workflows, further improvements in raw coding prowess will yield diminishing returns. To overcome these limitations, Windsurf argues, the next generation of SWE models must capture the full breadth and depth of the engineering process.

SWE-1

Windsurf claims SWE-1 to be the largest and most capable AI model in its lineup that performs competitively with Claude 3.5 Sonnet, GPT-4.1 and Gemini 2.5 Pro on internal programming benchmarks. However, SWE-1 appears to fall short of frontier AI models, such as Claude 3.7 Sonnet, in software-engineering tasks.

In a video announcing the SWE models, Windsurf’s head of research, Nicholas Moy, highlighted the company’s efforts to differentiate its approach. “Today’s frontier models are optimised for coding, and they’ve made massive strides over the last couple of years,” says Moy. “But they’re not enough for us … Coding is not software engineering.”

Windsurf notes that while other models excel at writing code, they struggle to navigate multiple surfaces such as terminals, IDEs and the internet as programmers often do. The startup states that SWE-1 was trained using a new data model and a “training recipe that encapsulates incomplete states, long-running tasks and multiple surfaces.”

The startup describes SWE-1 as its “initial proof of concept”, suggesting it may release additional AI models in the future.