Windsurf SWE-1
Windsurf's in-house software-engineering model family — trained from scratch for full-process development work, not just code output.
About
SWE-1 was Windsurf's first in-house model family, announced May 15, 2025: three variants (SWE-1 frontier, SWE-1-lite, SWE-1-mini) trained from scratch on a software-focused dataset rather than fine-tuned from an existing base. The pitch was full-process engineering — working across terminals, browsers, and IDEs with 'flow awareness' of unfinished tasks — instead of autocomplete-style code generation. Windsurf claimed near-frontier performance on internal evals, roughly comparable to Claude 3.5 Sonnet (provider-reported). The lite and mini variants were free for all Windsurf users; the frontier model was paywalled. Note: Windsurf was acquired by Cognition in 2025 and the SWE line continued under Cognition (SWE-1.5, 1.6, 1.7). One secondary outlet claimed an open license, but mainstream reporting describes the models as available only inside Windsurf — treat the open-license claim as unverified.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Comparable to Claude 3.5 Sonnet on internal evals — provider-reported
- SWE-L2 beat GPT-4 on SWE-bench code-reasoning per company-cited figures — provider-reported
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Launch covered by Gadgets360, TechEdt, and AllAboutAI (May 2025) — trained-from-scratch claim and three-variant lineup consistent across outlets; Performance figures are provider-reported internal evals only; no independent benchmarks in sources read; One secondary outlet claimed an open license but mainstream reporting describes Windsurf-only availability — flagged as unverified in entry.