Devin Adds GPT-6 Sol and Luna at Sharply Lower Cost

Cognition, the company behind the autonomous coding agent Devin (@cognition), has added OpenAI's newest models, GPT-6 Sol and GPT-6 Luna, as selectable options in Devin Desktop and CLI. Cognition tested both on FrontierCode 1.1, its own internal benchmark for whether a model's generated pull request is good enough to actually merge rather than just whether it passes automated tests, as described in Cognition's blog post announcing the change. On that benchmark, GPT-6 Sol matched the score of its predecessor, GPT-5.6 Sol, while costing 61% less per task, and GPT-6 Luna scored above GPT-5.6 Luna at about a quarter of the cost, putting it under $0.10 per task and, by Cognition's count, the cheapest model on its leaderboard.

These are Cognition's own numbers on its own benchmark, not an independent test, so they describe relative cost per completed coding task inside Devin rather than general model capability. The pattern matches what OpenAI itself reported the same day, that GPT-6 Sol and Luna carry roughly half the API price of the GPT-5.6 models they replace, and Artificial Analysis's separate measurement that the two models match their predecessors' Intelligence Index scores at about half the cost. For someone choosing a model for agentic coding work, the concrete result is a cheaper option that, on Cognition's own test, did not trade away code quality.