GPT-6.2 Astra and Sol: release, API pricing and Ultrafast
GPT-6.2 Astra and Sol are in ChatGPT Work, Codex and the API as of November 17. Astra costs $8 per million input tokens, 20% less. Let's see what it can do.
This article in other languages:Polski
In this article
On November 17, OpenAI released two new models: GPT-6.2 Astra and GPT-6.2 Sol. According to OpenAI’s announcement, both already work in ChatGPT Work and Codex on the Plus, Pro, Enterprise and Edu plans, and in the API they’re available as gpt-6.2-astra and gpt-6.2-sol.
What can GPT-6.2 Astra and Sol do?
Astra remains the model for the hardest projects. The post on the OpenAI forum reports new records on evaluations of professional work and scientific research, at a cost per task below one third of GPT-6 Astra’s. Presentations, websites and 3D designs are said to have improved too. 9to5Mac gives a concrete example: Astra takes a screenshot of a slide as a template and reproduces its layout, colors and font size.
Sol is the model for everyday work and, as the forum post puts it, the main model for many OpenAI employees. The more interesting number comes from Cognition’s FrontierCode 2.0 benchmark, which checks whether changes to real repositories are fit to merge. According to OpenAI, Sol at maximum reasoning effort trails Astra by half a point and costs about one fifth as much (these are OpenAI’s measurements, so let’s take them with a grain of salt).
Both models are meant to finish tasks using fewer output tokens. In ChatGPT and Codex that means more work before hitting the limit, and in the API a lower bill, whatever the price list says.
How much does GPT-6.2 cost in the API?
Prices and limits come from the model pages for GPT-6.2 Astra and GPT-6.2 Sol (as of November 17, 2026, rates per million tokens):
| GPT-6.2 Astra | GPT-6.2 Sol | |
|---|---|---|
| Input | $8 | $2 |
| Cached input | $0.40 | $0.10 |
| Output | $40 | $10 |
| Context window | 1,050,000 tokens | 1,050,000 tokens |
| Max output | 128,000 tokens | 128,000 tokens |
| Knowledge cutoff | May 18, 2026 | June 1, 2026 |
Astra’s standard prices are 20% lower than GPT-6 Astra’s, and Sol keeps the GPT-6.1 price list. Interestingly, $8 and $40 is exactly what Claude Fable 5.2 has cost since November 9, so the flagship models from OpenAI and Anthropic now share the same price list.
The table doesn’t tell the whole story. The Sol page lists a long-context surcharge: on a request with more than 272K input tokens, input and cache rates double and the output rate goes up by half, and that applies to the entire request. So the million-token window is there, but not at the price in the table.
Tasks that can wait come out cheaper, though. Batch for Sol costs half the standard rate, and the API changelog from the same day adds background completion windows (in a public beta of the Responses API): a two-hour window with rates 30% lower and an eight-hour window with rates 40% lower.
Ultrafast on the $200 plan
Ultrafast, the mode that generates tokens up to 4.5 times faster, used to be out of reach for the $200-a-month Pro plan. Now its subscribers get it for the first time, with Sol only for the time being. The $500 Pro plan has Ultrafast with both models, and in Enterprise and Edu an administrator has to turn it on.
Speed has its price. In ChatGPT, this mode burns through the limit that ChatGPT and Codex are set to share from December faster. In the API, Ultrafast for Sol costs three times the standard rate, but only until February 28, 2027. From March 1 it will be six times.
What should we watch out for?
First, the API migration. Both models support Chat Completions and the Responses API, but according to the changelog tool calling requires the Responses API. If our agent uses tools through Chat Completions, swapping the model name won’t be enough.
Second, reasoning effort. The API rejects requests that set it to none, so these models can’t be run without reasoning.
Third, what the announcement leaves out. In the Hacker News discussion, where the release hit the top spot within minutes, commenters pointed out that the link to the system card didn’t work at launch, and that the charts from the science benchmarks don’t include results for Gemini 4 Carbon (I wrote about that model in the Gemini 4 Carbon article).
FAQ
When was GPT-6.2 released?
GPT-6.2 Astra and Sol have been available since November 17, 2026 in ChatGPT Work, Codex and the API.
Is GPT-6.2 available for free?
The announcement lists the Plus, Pro, Enterprise and Edu plans. The free plan isn’t on that list.
What’s the difference between GPT-6.2 Astra and Sol?
Astra is the model for the hardest projects, and Sol is for everyday work. In the API, Sol costs a quarter as much per token ($2 and $10 versus $8 and $40 per million tokens), with the same context window and output limit.