GPT-6.2 is here: what changes for ChatGPT users?
OpenAI released GPT-6.2 in two versions, Astra and Sol. Astra is 20 percent cheaper through the API, Sol keeps its price. What it costs, who gets it and what your business can do with it.
Lees in het Nederlands →
Short answer: OpenAI released GPT-6.2 on 17 November 2026 in two versions: Astra for heavy thinking and Sol for everyday work. Through the API, Astra is 20 percent cheaper than its predecessor, while Sol keeps the same price. In ChatGPT you'll see them as a preview in Work and Codex, on the paid plans.
Every few months a new model arrives, and every time the question is the same: do I need to change anything? With GPT-6.2 the answer for most companies is "a little". In this article I list what's new, what it costs and what you can do with it. Status as of 21 November 2026.
What is GPT-6.2?
GPT-6.2 is OpenAI's newest generation of models. It comes in two flavours:
- GPT-6.2 Astra is, according to OpenAI, the new frontier model "for your hardest projects": research, software and other complex professional work.
- GPT-6.2 Sol is the efficient, versatile model for everyday work, with strong reasoning at a higher speed.
Both models are available in ChatGPT Work, in Codex and in the API.
What does GPT-6.2 cost through the API?
Prices per million tokens, according to OpenAI's announcement:
- GPT-6.2 Astra: 8 dollars input, 0.40 dollars cached input, 40 dollars output. That is 20 percent less than GPT-6 Astra.
- GPT-6.2 Sol: 2 dollars input, 0.10 dollars cached input, 10 dollars output. That is the same price as GPT-6.1 Sol.
In practice, Astra is four times as expensive as Sol. For a chatbot on your website or summarising emails you rarely need it. How to keep your usage under control is explained in my guide to topping up your OpenAI account.
What changes in ChatGPT?
OpenAI's announcement mentions the Plus, Pro, Enterprise and Edu plans. There you choose the models in the Work model picker or in Codex. According to OpenAI's release notes, it is a preview for now. Free users are not mentioned.
A few things to know:
- Business: OpenAI's announcement doesn't explicitly mention the Business plan, and other media contradict each other on this. So check your own workspace.
- Enterprise and Edu: the admin has to enable the models first.
- Ultrafast: a faster mode that uses more of your limit. For now it is available on Pro 500 and through the API.
- Limits: according to OpenAI's pricing page, from December you get one shared limit for ChatGPT and Codex together.
Is GPT-6.2 better than Claude?
It depends on who you ask. On Artificial Analysis, an independent ranking, Astra scores 62 and Sol 57, against 65 for Anthropic's Claude Fable 5.2 (as reported by Decrypt and WinBuzzer). That puts Astra in second place.
What I find striking is in OpenAI's own safety report. In two tests, both models more often kept quiet about an obstacle they knew about. For Astra that happened in about 22 percent of the trials, against 11 percent for GPT-6 Astra (figures from the system card, reported by WinBuzzer). If you let GPT-6.2 work as an agent, check what it says it has done.
And Microsoft's cheaper models?
On the same day, 17 November 2026, Microsoft presented a series of its own MAI models at Ignite. For teams that have a lot of code written, MAI Code 2 is the most interesting: according to Microsoft it reaches 98 percent of the coding quality of GPT-6.1 Sol at about 40 percent of the price. You can already try it in GitHub Copilot. MAI Image 2.7 for images is available in Microsoft Foundry.
Note: those comparisons come from Microsoft itself, and the announcement gives no concrete per-token prices. If your team already uses Copilot, also read my article on the new Copilot Chat limits.
What should you do now?
- Do you use ChatGPT Plus or Pro? Try Sol for your daily tasks and Astra only for really hard work.
- Do you use the API? Compare your current bill with the new Astra price. If you already run on Sol, your costs don't change.
- Do you work with agents? Have them report on important tasks and check that report.
- Keep your API keys safe when you build new connections. My explanation of what an API key is helps with that.
My opinion?
Astra's price cut is real and welcome. Still, I think most companies will notice little of it, because Sol is enough for ordinary work and its price doesn't change.
What I actually find most interesting is the honest bit in the safety report: a smarter model isn't always a more honest model. So don't automatically pick the most expensive one, pick the model that fits the task, and keep checking what it does.
Frequently asked questions
Can I use GPT-6.2 for free?
Not according to OpenAI's announcement. The models are on the paid Plus, Pro, Enterprise and Edu plans, for now as a preview.
What is the difference between Astra and Sol?
Astra is the heavy model for complex work and costs 8 dollars input and 40 dollars output per million tokens through the API. Sol is faster and more efficient, for everyday tasks, at 2 and 10 dollars.


