GPT-6 Sol and Luna Cut ChatGPT Costs in Half

OpenAI has just made its most capable everyday model roughly half the price it was a month ago, and that changes the maths for any business running ChatGPT at scale.

What happened

On 22/09/2026 OpenAI released two new models, GPT-6 Sol and GPT-6 Luna, both trained using the same methods as GPT-6 Astra, the computer use model it launched earlier this month. Sol is the more capable of the pair, built for harder reasoning and coding work. Luna is the value option, tuned for speed and cost.

The headline is pricing. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens. GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens. Both figures are roughly 50 per cent lower than GPT-5.6 promotional pricing. OpenAI also says Sol halves the factual error rate of GPT-5.6 Sol, and that Luna at higher effort settings matches GPT-5.6 Sol's output while costing around a hundredth as much.

Access rolled out gradually through the day. Plus, Pro, Business, Enterprise and Edu account holders got both models in ChatGPT Work and in Codex from day one. Free and Go users only get Luna, and only through the desktop app rather than the browser. Developers can call the models through the API as gpt-6-sol and gpt-6-luna, and both benefit from an improved prompt caching system that discounts repeated input tokens by up to 90 per cent.

Why it matters for UK businesses

Most small and mid sized firms using ChatGPT are not running frontier research workloads. They are drafting emails, summarising documents, building spreadsheets and answering customer queries, which is exactly the kind of everyday work Sol and Luna are built for. A halving of API costs matters most to anyone who has built ChatGPT into a product or an internal tool rather than just using the chat window, because token costs compound fast once a workflow runs automatically rather than being typed by a person.

The Free and Go restriction is worth noting too. If your team is on the lower tiers and relies on the browser rather than the desktop app, you will not see Luna yet, only the older model. That is an easy thing to miss when everyone assumes an upgrade lands everywhere at once.

How Adevious AI sees it

We think this is a genuine step change for anyone running ChatGPT through the API rather than the chat interface, and we would move automated workflows, especially high volume ones like ticket triage or first draft generation, onto Luna first to test whether quality holds at that price point before touching anything that faces a customer directly.

The caution is that benchmark wins on release day do not always survive contact with real, messy business data. OpenAI's own comparisons lean heavily on automated coding and agent benchmarks, not on the kind of varied, occasionally scruffy documents a UK small business actually feeds a model. Before you swap a production workflow onto Sol or Luna, run your own side by side test on real examples from your business, not just the demo prompts, and watch output quality for at least a week before trusting it unsupervised.

One practical takeaway this week

If you or your team use the ChatGPT API, or a tool built on top of it, check which model your integration is calling by default. Many tools default to whatever was current when they were built, not the newest release, so a manual switch to gpt-6-luna or gpt-6-sol may be sitting there unclaimed. That alone could cut your monthly bill without changing a single prompt.

If you are on Free or Go and use the browser rather than the desktop app, do not assume you already have the new model. Check before you plan around it.

If you want a second opinion on whether your AI setup is using the right model for the job, or you would like help testing GPT-6 Sol and Luna against your own workflows before switching, get in touch with Adevious AI. We are happy to talk it through.

Sources: https://openai.com/index/introducing-gpt-6-sol-and-luna/ https://openai.com/index/better-prompt-caching-for-gpt-6/ https://x.com/OpenAI/status/2102460995180663204

Acronyms: API: Application Programming Interface UK: United Kingdom