OpenAI launched GPT-6 Sol and GPT-6 Luna on September 22, 2026, with API access and a rollout in ChatGPT Work and Codex. The two models extend the GPT-6 family to work where the cost of repeated requests matters as much as the strongest possible answer.
Sol is positioned for complex coding and agent workflows. Luna targets focused tasks at high volume. Both accept text and images and produce text, so image input means understanding visual material, not generating a picture.
Where You Can Access Them
The announced paid-plan access covers Plus, Pro, Business, Enterprise and Edu. Free and Go users can access Luna in the desktop app. The launch covers Work and Codex, with ordinary Chat excluded. Developers can call the models as gpt-6-sol and gpt-6-luna.
The timing matters to me. On September 10, 2026, new sign-ups and upgrades to $200 Pro were paused, while existing subscriptions continued. Tibo Sottiaux, who works on Codex and ChatGPT at OpenAI, said this tier put the most strain on its systems while the company added capacity. Capability had become an access problem as well as a performance question.
What the Lower Prices Cover
Standard API text-token rates for Sol are $2 per million input tokens, $0.20 for cached-input reads and $10 for output. Luna’s rates are $0.10, $0.01 and $0.50 respectively. These apply to prompts up to 272,000 input tokens; cache writes, longer prompts and other processing modes have separate pricing.
Sol’s input and output rates are half its GPT-5.6 promotional prices. They are also one-fifth of Astra’s $10 input and $50 output rates. Those are API prices, not subscription prices or a promised reduction in the cost of completing a task.
The same-day caching update reuses computation across requests. That interests me for an agent carrying the same brief and sources into successive turns: repeated context can become a recurring expense even when the new question is small.
OpenAI reports that Sol scored 68.8% on DeepSWE 1.1 at maximum reasoning effort, while Luna scored 66.6%. The evaluation covers extended software-engineering tasks. These are published launch results, and the research or API setup can differ from production ChatGPT.
What do I think of this?
I think Sol and Luna make the GPT-6 family more practical. Astra’s depth could matter for a difficult investigation; a narrow extraction or routine follow-up may not need that same capability. Less expensive options could leave more room for iteration within a fixed budget.
The Pro pause makes sustained access part of the value question for me. I would welcome cheaper models that can carry routine steps reliably while Astra remains useful for deeper work. That could make longer assignments more affordable, although token prices alone do not reveal the computing capacity they need.
For me, the convincing next evidence would be comparable finished tasks: the same brief, the same tools, the resulting output and the total consumption. A cheaper answer that needs several retries could erase the savings.