SNACK: 3-line summary
- OpenAI announced GPT-6 Sol and GPT-6 Luna on September 22, 2026.
- API prices are 50% below GPT-5.6 promotional pricing, with Sol aimed at demanding work and Luna at routine, high-volume tasks.
- ChatGPT Work and Codex access is rolling out gradually to Plus, Pro, Business, Enterprise and Edu users; neither model is available in Chat yet.
OpenAI announced GPT-6 Sol and GPT-6 Luna on September 22 with API prices 50% lower than GPT-5.6 promotional pricing. The company attributes the reduction to improved caching and inference, while giving developers two distinct options for demanding projects and repeatable everyday work.
Snackgirls react
AIKO: The price cut makes me want to compare costs per finished task. Then I’d open the cache dashboard: repeated context is a less charming habit when you keep paying to process it.
Red: I want to give Sol a long Codex task with several connected changes. I’d be watching whether its final report matches what it actually finished.
Sol for difficult projects, Luna for repeatable tasks
OpenAI positions Sol for difficult professional and coding work. Luna is the lower-cost option for high-volume requests with clear goals, such as summarization, extraction and quick answers. That gives developers a practical starting point for routing tasks instead of sending every request to the same model.
OpenAI also reports that Sol made about half as many mistakes as GPT-5.6 Sol in an internal factuality evaluation based on de-identified conversations where users flagged errors. It says both new models make fewer misleading claims about completed coding work than their GPT-5.6 counterparts. These are vendor-reported results, not independent proof of reliability on your workload.
Where you can select the models
According to OpenAI, both models are rolling out gradually in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, and are also available through the API. Neither model is available in Chat yet.
Free and Go users can access GPT-6 Luna in the desktop app. If you use ChatGPT Work or Codex on one of the paid plans listed above, check that service’s model picker; if the new models are not visible, try again later.
Measure completed-task cost and cache misses
Lower API prices change the cost of requests, but the cost of finishing a task still depends on token usage, retries and the quality of the result. When comparing Sol and Luna, track those factors on representative jobs rather than relying on the price reduction alone.
The companion prompt-caching update adds a dashboard and diagnostics tool that can reveal cache misses. Eligible shared prefixes reused within a 30-minute window can receive cached-input discounts of up to 90%. For applications that repeatedly send the same instructions or context, checking whether those requests actually hit the cache is a concrete place to start.
OpenAI also says GPT-6 applications can change reasoning effort between responses without breaking the cache when they use the documented configuration update path. If your application varies that setting, check the update path alongside the new cache diagnostics.
Sources and date: 2026-09-23 KST