OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, two cheaper tiers below its top model, GPT-6 Astra. In the API, Sol costs $2 and Luna $0.10 per million input tokens, half of GPT-5.6’s current prices. Regular ChatGPT chats do not use them yet.
New to this? Read it in simple words
- OpenAI released two new AI models, GPT-6 Sol and GPT-6 Luna, on September 22.
- Sol is made for coding and agent work. Luna is made for simple tasks in large numbers.
- The big change is the price. They cost about half as much as the models they replace.
- The first independent test found them about as smart as the older models, not smarter.
- Token
- A small piece of text, often part of a word. AI use is priced per token.
- AI agent
- An AI that takes steps on its own to finish a task, such as booking or coding.
Half of today’s price, after earlier cuts
In OpenAI’s API, GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens. GPT-6 Luna costs $0.10 and $0.50. A token is a small piece of text, often part of a word.
The models they replace are GPT-5.6 Sol, at $4 and $20, and GPT-5.6 Luna, at $0.20 and $1.20. So the new prices are about half. But OpenAI had already cut the GPT-5.6 prices in July and August, so the drop from their July launch prices is much larger.
An OpenAI spokesperson told VentureBeat that the new prices are permanent. Text that the model reads again from its cache costs 90 percent less. Prompts longer than 272,000 tokens cost more.
Prices are per million input and output tokens. The comparison is with GPT-5.6’s current prices, which OpenAI had already cut.
Developers first, regular chats later
Developers can use the models now under the names gpt-6-sol and gpt-6-luna. GitHub is adding both to Copilot step by step. Microsoft also offers them to business customers in its Foundry cloud service.
In ChatGPT, paid plans get both models in ChatGPT Work and Codex, with a gradual rollout. Free and Go users get only Luna, and only in the desktop app.
Regular ChatGPT chats do not use the new models yet. Paid users still chat with GPT-5.6 Sol, and OpenAI has not said when that will change.
OpenAI’s claims, and the first outside test
OpenAI says Sol makes about half as many mistakes as the model it replaces. It measured this on chats where users had flagged errors, and it says those chats are not typical.
In OpenAI’s own tables, Sol scores 33.2 percent on AutomationBench, ahead of Anthropic’s Claude Opus 5 at 26.9 percent. It is slightly behind Claude Fable 5 on the DeepSWE coding test. OpenAI compared Sol with the older Opus 5, not with Opus 5.5, which Anthropic released about 90 minutes earlier.
Artificial Analysis, an independent testing company, gives Sol 48 points and Luna 37 on its Intelligence Index. Astra scores 53. It finds the new models about level with GPT-5.6, at roughly half the cost per task.
OpenAI’s system card puts both models at its High level for cybersecurity and for biology and chemistry. That is one step below Astra, which OpenAI rates Critical for cyber. Sol and Luna use the same safeguards as GPT-5.6.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3
- 4
- 5Research · September 22, 2026OpenAI releases GPT-6 Sol and Luna models, slashing API costs 50% or more VentureBeat
- 6Research · September 22, 2026OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes TechCrunch
- 7Research · September 22, 2026GPT-6 Sol and Luna push the cost efficiency frontier Artificial Analysis
- 8
- 9Primary source · September 22, 2026GPT-6 Astra, Sol, and Luna: For production agents in Microsoft Foundry Microsoft Azure
We read OpenAI’s launch post, API pricing page, ChatGPT release notes, and system card, plus the GitHub and Microsoft announcements. We compared them with VentureBeat, TechCrunch, and the independent tests from Artificial Analysis. Quality and safety figures come from OpenAI, except where we name Artificial Analysis.