OpenAI's two new workhorse models cut the cost of running AI by about half. The bigger story for your firm is where buyers now go for their answers.
On 22 September 2026, OpenAI added two models to its GPT-6 line: GPT-6 Sol and GPT-6 Luna. They sit below the GPT-6 Astra flagship that arrived on 3 September, so the family now runs across three tiers. The company says the pair costs roughly 50% or more less to run than the GPT-5.6 tiers they replace. OpenAI confirmed the release on its own announcement page, and a spokesperson told VentureBeat the lower rates are "permanent prices, not promotional or introductory pricing".
OpenAI fills in the cheaper end of GPT-6
The two models do different jobs. Sol is the everyday workhorse for demanding, repeated knowledge and development work: building features, reviewing code, debugging and analysing data. Luna is cheaper again and built for high-volume, tightly defined tasks such as summarising documents, pulling out specific facts and answering straightforward questions. Both read text and images and reply in text.
The prices are the headline. In the API, Sol costs $2 per million input tokens and $10 per million output tokens. Luna costs $0.10 per million input tokens and $0.50 per million output tokens. For reference, the Astra flagship runs at $10 and $50 for the same. Those figures come from VentureBeat's reporting, and OpenAI says they are permanent rather than an introductory discount.
Access differs by product. Developers can use both models through the API now, under the IDs gpt-6-sol and gpt-6-luna. Inside ChatGPT Work and Codex, the models are rolling out gradually to Plus, Pro, Business and Enterprise, with no single completion date given. Free and Go users can reach Luna through the desktop app. General chat access was not available when OpenAI announced the release.
Why cheaper AI widens where buyers find you
The interesting part is not the price list. It is what a cheaper "answer these questions" model does to the market. Summarising, extracting and answering plain questions is exactly the work Luna is priced for, and it is exactly the work behind the AI assistants your clients already use when they shortlist advisers. When that work gets cheaper, it tends to get used more, in more products, on more queries.
For a professional-services firm, none of this is a lever you pull. There is no ranking factor here and no setting to change on your website. Being named in AI answers, and getting recognised by ChatGPT and the assistants like it, is still an emerging area rather than a settled part of search. The measured response is the one that has held all year: keep the public facts about your firm clear, accurate and easy to quote, so that when an assistant fields a buyer's question about firms like yours, it has something correct to say about you. That is the ground on which earning AI recommendations and your firm's AI visibility are built.
Frequently asked questions
Do I need to do anything because of this release?
No. This is a pricing and model update for developers and ChatGPT users, not a change to search rankings or a setting on your website. There is nothing to switch on. The sensible response is to keep the public facts about your firm accurate and easy to quote.
What is the difference between GPT-6 Sol and GPT-6 Luna?
Sol is the everyday workhorse for demanding, repeated knowledge and development work such as building features, reviewing code and analysing data. Luna is cheaper again and built for high-volume, simple jobs like summarising, extracting facts and answering straightforward questions. Both handle text and images and reply in text.
How much cheaper are they?
OpenAI says the pair costs roughly 50% or more less to run than the GPT-5.6 tiers. In the API, Luna is priced at $0.10 per million input tokens and $0.50 per million output tokens, and Sol at $2 and $10. An OpenAI spokesperson told VentureBeat the lower rates are permanent, not a temporary promotion.
Can my team use these models today?
Developers can use both through the API now. Inside ChatGPT Work and Codex the models are rolling out gradually to Plus, Pro, Business and Enterprise, and Free and Go users can reach Luna through the desktop app. General chat access was not available when OpenAI announced the release.
Will this affect whether buyers find my firm?
Not directly, and not through rankings. Cheaper models make AI assistants cheaper to run, so they tend to answer more questions in more places. Being named in those answers is still an emerging area rather than a settled ranking factor, so the useful step is to keep clear, accurate, quotable pages about your firm.
