Back to rankings

Windsurf: GPT-6.1 Sol Now Available in Devin Desktop and CLI

WindsurfView original changelog

Windsurf's Devin now offers GPT-6.1 Sol in Devin Desktop and the Devin CLI. The model matches GPT-6 Sol's 60.4% FrontierCode 1.1 score while costing 44% to 57% less across reasoning effort levels. At medium effort a task costs about $0.31, down from $1.66, and the low effort setting posts the highest score of any model under $0.30 per task on the leaderboard.

Key Takeaways

  • GPT-6.1 Sol matches GPT-6 Sol's 60.4% FrontierCode 1.1 score, so the upgrade is about efficiency rather than a jump in raw capability.
  • Medium effort tasks now cost about $0.31 instead of $1.66, an 81% reduction that makes heavy agent usage far more affordable.
  • Across effort levels the model is 44% to 57% cheaper while matching or beating its predecessor.
  • The low effort setting climbed from 50.5% to 58.1% at around $0.21 per task, the highest score under $0.30 per task on the leaderboard according to Cognition.
  • The model ships in both Devin Desktop and the Devin CLI, so it is usable from the editor and the terminal on day one.
  • Cheaper strong models make it practical to run more parallel or long-running agent sessions without exhausting usage limits.

GPT-6.1 Sol Arrives in Devin

Windsurf's Devin made GPT-6.1 Sol available on September 29, 2026, in both Devin Desktop and the Devin CLI. The release is a cost-efficiency upgrade over the earlier GPT-6 Sol: it holds the same top-line quality on Cognition's FrontierCode 1.1 benchmark while making each task considerably cheaper to run. Developers who already pick GPT models inside Devin can switch to the new version from the model picker with no other changes to their workflow.

Same Quality, Much Lower Cost

On FrontierCode 1.1, GPT-6.1 Sol scores 60.4%, matching the prior version. The difference is in what it takes to get there. At medium reasoning effort, the average cost per task is $0.31, an 81% reduction from the $1.66 that GPT-6 Sol required. Across all effort levels, Cognition reports the new model is 44% to 57% cheaper while matching or exceeding GPT-6 Sol's performance.

A Stronger Low-Effort Option

The low effort setting improved the most. GPT-6.1 Sol now scores 58.1% at roughly $0.21 per task, up from 50.5% for the previous model at that setting. According to Cognition, that is the highest score of any model under $0.30 per task on the FrontierCode leaderboard. For everyday edits, quick fixes, and high-volume agent runs, this makes the low setting a far more credible default than before.

How to Try It

GPT-6.1 Sol is available now through the Devin Desktop download and by installing the Devin CLI. Because it targets cost rather than raw capability, it is most relevant to people who run many agent sessions and want to stretch their usage allowance further without giving up quality.