GPT-6 Sol and Luna put everyday AI workloads in focus
The useful choice is not simply the strongest model, but the least expensive one that reliably completes the job.
AI · Macfolio · 22 September 2026
OpenAI's September 22 release of GPT-6 Sol and Luna extends the family below Astra, with lower API prices and improvements aimed at everyday work.
Sol costs $2 per million input tokens and $10 per million output tokens. Luna costs $0.10 and $0.50 respectively. Those are token rates, not fixed prices for a finished task: the amount of context, generated output, and repeated work still matters.
Match the model to the task
Our take is to separate quick, bounded work from ambiguous work. A short classification or extraction task is a different test from changing a complex codebase. Evaluate accuracy and completion time on your own examples before deciding that a faster response is a better result.
Availability also needs checking. The launch covers ChatGPT Work, Codex, and the API; OpenAI says these models are not yet in Chat. Benchmark improvements are the company's measurements, not a guarantee for every workflow.