On November 17, 2026, OpenAI released two new models: GPT-6.2 Astra and GPT-6.2 Sol. Astra is the flagship, built for long and difficult jobs. Sol is the one for everyday work and, according to OpenAI, it costs far less for similar results. Both are already available in ChatGPT Work, in Codex and through the API.
The most concrete news is the price. Astra costs 20% less than its predecessor, GPT-6 Astra.
What is the difference between Astra and Sol?
OpenAI describes Astra as its most capable model, meant to carry a project from start to finish. Sol is the default model for daily work, coding and agents (programs that handle a series of steps on their own, such as opening files, searching the web and filling in a document).
On paper the two models are very similar. Here are the technical specs from the developer documentation:
| Spec | GPT-6.2 Astra | GPT-6.2 Sol |
|---|---|---|
| Maximum context | 1,050,000 tokens | 1,050,000 tokens |
| Maximum output | 128,000 tokens | 128,000 tokens |
| Knowledge cutoff | May 18, 2026 | June 1, 2026 |
| What it reads | Text and images | Text and images |
| What it produces | Text only | Text only |
A token is a small piece of a word. A million tokens is, roughly, several hundred pages: enough to hand the model a whole folder of contracts or a full manual in one go. Neither Astra nor Sol reads audio or video.
How much do they cost through the API?
These are the standard prices in dollars per million tokens, taken from the official pricing page, compared with GPT-6:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Astra | 8.00 | 0.40 | 40.00 |
| GPT-6 Astra | 10.00 | 1.00 | 50.00 |
| GPT-6.2 Sol | 2.00 | 0.10 | 10.00 |
| GPT-6 Sol | 2.00 | 0.20 | 10.00 |
"Cached input" is text you send again unchanged, for example the same instructions or the same reference document: it costs a fraction of the normal price. On Sol that rate has been cut in half, while the rest stays the same.
Before you do the math, keep a few rules in mind:
- above 272,000 input tokens the price goes up: for Astra, input rises to 16 dollars and output to 60;
- Ultrafast mode, the faster option, costs three times the standard price until February 28, 2027, and six times from March 1, 2027;
- requests that can wait get a discount: 30% if you accept an answer within 2 hours, 40% within 8 hours.
What does OpenAI claim about performance?
All the numbers below come from OpenAI, and nobody has checked them independently yet. According to the company, Astra completes a task at less than a third of the cost of GPT-6 Astra. On Agents' Last Exam, a test of how well a model finishes complex jobs on its own, Astra beats GPT-6 Astra by 5.2 points while spending less than half.
For Sol, OpenAI says it comes close to Astra on the FrontierCode 2.0 coding test, at about a fifth of the cost. That is the most interesting figure in the announcement, because it suggests the cheaper model is enough for many jobs.
OpenAI also talks about cybersecurity: Astra is the strongest model it has ever had in this field, so the company has added stricter safeguards on risky requests.
What changes for people at work?
If you use ChatGPT on a Plus, Pro, Enterprise or Edu plan, the announcement on OpenAI's developer forum says you will find the new models in ChatGPT Work and Codex. Availability depends on your plan, your workspace permissions and how far the rollout has gone, so you may not see them right away. The announcement says nothing about the free plan.
The change matters most if you pay per use, as companies do when they connect GPT to spreadsheets, business software or automations. If you use Astra, you spend a fifth less right away by changing only the model name. If you often resend the same documents, you also save on Sol, thanks to the halved cache price.
Our advice: start with Sol. For summaries, emails, table analysis and first drafts, the pricier model rarely makes a difference, and Sol costs a quarter as much. Move to Astra only for the jobs where Sol keeps getting things wrong. If you are still choosing an assistant, we compare them in ChatGPT, Claude or Gemini: which one to choose for work.
Frequently asked questions
What are the model names in the API?
gpt-6.2-astra and gpt-6.2-sol. They work with the Chat Completions, Responses and Batch interfaces. You cannot train them on your own data (fine-tuning).
What is Ultrafast on the Pro plans?
It is a mode that answers faster. According to OpenAI, in ChatGPT the 200-dollar Pro plan gets it with Sol, while the 500-dollar Pro plan gets it with both models.
Should you update your automations now?
If you use GPT-6 Astra, yes, after a test on a few real cases: the price drops and the model is newer. If you use GPT-6 Sol, you only save on the cached part, so you can update when it suits you.
Official sources
- OpenAI: Introducing GPT-6.2 Astra and Sol
- OpenAI Developer Community: availability announcement
- OpenAI: API pricing
- OpenAI: GPT-6.2 Astra model page
- OpenAI: GPT-6.2 Sol model page
This article was written with the help of artificial intelligence models, starting from the official sources listed, and is reviewed before publication.














