Windsurf has unveiled its first household of specialised fashions designed to remodel builders’ work in a major improvement for AI-assisted software program engineering. The SWE-1 household, introduced on Might 15, 2025, represents a elementary shift in AI help for builders, transferring past mere code era to embody the complete software program engineering workflow.
Past Simply Writing Code
Whereas latest years have seen outstanding enhancements in AI coding capabilities, Windsurf acknowledged a vital limitation in present approaches: Software program improvement includes far more than simply writing code.
“Why construct SWE-1? Merely put, our aim is to speed up software program improvement by 99%. Writing code is just a fraction of what you do. A ‘coding-capable’ mannequin gained’t minimize it,” explains the Windsurf staff of their announcement.
The SWE-1 household consists of three distinct fashions:
- SWE-1: The flagship mannequin, corresponding to Claude 3.5 Sonnet in tool-call reasoning capabilities, whereas being extra cost-efficient to run. It’s briefly accessible to paid customers at zero credit per immediate.
- SWE-1-lite: A mid-sized model that replaces Windsurf’s earlier Cascade Base mannequin, providing all customers improved high quality and limitless use.
- SWE-1-mini: A small, ultra-fast mannequin powering Windsurf Tab’s passive expertise for all customers.
Move Consciousness: The Key Innovation
What units Windsurf’s strategy aside is its idea of “movement consciousness”—the power for AI techniques to know and function inside the full, shared timeline of improvement work. This perception got here from the corporate’s well-liked Windsurf Editor, which permits seamless collaboration between people and AI.
This movement consciousness permits the fashions to know incomplete work states and swap naturally between AI and human contributions. If a mannequin makes an error, the human can soar in to right it, and the mannequin can then proceed working based mostly on these corrections, creating a really collaborative workflow.
“It will likely be some time earlier than any SWE mannequin can actually do every thing independently,” acknowledges Windsurf. “Move consciousness permits the best type of interplay throughout this intermediate interval.”
Spectacular Benchmark Efficiency
In response to Windsurf’s analysis knowledge, SWE-1 performs comparably to frontier fashions from main AI labs and considerably outperforms mid-sized and open-weight options. The corporate makes use of two major benchmarks:
- Conversational SWE Process Benchmark: Testing how nicely a mannequin can handle the next consumer question in the midst of an present session with a half-finished process.
- Finish-to-Finish SWE Process Benchmark: Evaluating a mannequin’s capacity to resolve an issue independently of starting to finish.
In manufacturing experiments with actual customers, SWE-1 demonstrated robust efficiency in metrics like “Each day Traces Contributed per Person” and “Cascade Contribution Price,” reflecting each the standard of its ideas and customers’ willingness to undertake them.
A DevOps Perspective
These developments maintain specific promise for DevOps professionals. The SWE-1 fashions’ capacity to work throughout a number of surfaces—together with the terminal, textual content editor, and browser—aligns completely with the built-in nature of recent DevOps workflows.
The fashions can:
- Incorporate terminal outputs and perceive errors
- Seamlessly transition between textual content enhancing and debugging
- Keep consciousness of terminal instructions and IDE actions
- Course of consumer suggestions and testing outcomes
These capabilities might considerably streamline the usually advanced handoffs between improvement and operations phases that DevOps groups handle each day.
“Windsurf’s SWE-1 announcement is a transparent indicator that the way forward for software program improvement is quickly changing into AI-driven, extending far past easy code era,” stated Mitch Ashley, VP and observe lead, DevOps and utility improvement at Futurum. “I applauded Windsurf’s ambition to handle the complete improvement course of with AI, integrating human-AI interplay at a elementary degree. This raises the bar for all distributors within the area, pushing them to ship extra holistic, contextually conscious, and actually agentic capabilities to builders.”
Constructing a Software program Engineering Flywheel
Windsurf’s strategy represents a promising flywheel impact: As customers work together with their instruments, the corporate good points worthwhile insights into the place fashions want enchancment, enabling it to reinforce mannequin capabilities repeatedly.
“We at all times know, at scale, precisely what our customers need us to enhance with our fashions subsequent,” notes the Windsurf staff. “That’s how we’ve quickly constructed our mannequin to the extent it has achieved in right now’s SWE-1 state.”
What’s Subsequent for Windsurf
Whereas the corporate is happy with its preliminary outcomes, it emphasizes that SWE-1 is only the start. Windsurf plans to speculate considerably additional in its mannequin improvement, with the bold aim of not simply matching however exceeding the efficiency of frontier fashions from main analysis labs inside the software program engineering area.
For the rising variety of DevOps groups integrating AI instruments into their workflows, Windsurf’s deal with the whole engineering course of somewhat than simply coding duties represents a promising evolution that would assist bridge the normal gaps between improvement and operations.
As software program groups proceed exploring how AI can improve their productiveness with out sacrificing high quality or maintainability, Windsurf’s “flow-aware” strategy gives an intriguing mannequin for human-AI collaboration that respects the advanced, iterative nature of recent software program improvement.
