OpenAI: GPT-6.1 Sol is in Work, not in Chat yet
OpenAI presents GPT-6.1 Sol as a more capable Sol, close to Astra on some work, at one fifth of Astra’s listed price. It is in ChatGPT Work and Codex. It is not in Chat yet.
Getting started
Original AI-generated illustration · SecuFocus
At a glance
Key points
If you use the public chat, nothing changes today. If you are on a Work, Pro, Business, Enterprise or Edu account, or if you call the API, the model can already drive tasks and applications. Each question leaves for OpenAI. The scores on the page are the publisher’s, not a SecuFocus test.
What changes
On 29 September 2026, OpenAI’s product list dates the GPT-6.1 Sol announcement. The announcement page describes it as an upgrade to GPT-6 Sol, close to GPT-6 Astra on agentic coding, computer use and professional work, at one fifth of Astra’s listed input and output prices.
The API name is gpt-6.1-sol. The announced prices are 2 dollars per million input tokens, 0.10 dollar for cached input, and 10 dollars per million output tokens. OpenAI also says an Ultrafast variant is coming, up to eight times faster at generating tokens in Codex, in the following days. That variant was not the product available on announcement day.
| Point | What the announcement says |
|---|---|
| Where | ChatGPT Work and Codex, for Plus, Pro, Business, Enterprise and Edu |
| Where not | Not in Chat yet |
| API | gpt-6.1-sol |
| Announced price | $2 / $0.10 cached / $10 per million tokens |
Who it is for
The useful split is between chat and work. If your use is the public conversation window, the announcement says the model is not there yet. If your account is Work, or a tool calls the OpenAI API, this may already be the model that answers.
The sensitive point is not the name. It is computer use. The announcement describes tasks where the model interacts with applications. Giving that kind of tool means letting it act within what the account and the approvals allow. That is no longer only a reply on a page.
What leaves
GPT-6.1 Sol is not a model you run at home. A question sent to ChatGPT Work, Codex or the API leaves your machine for OpenAI. A cheap cache does not change that direction. It only says that reusing a context you already sent costs less on the next call.
The announcement points to a system-card addendum for the safety evaluation detail. This note does not absorb that addendum. It also does not turn the announcement’s tables into an independent measurement. OpenAI says its evaluations were run in its research environment or through its API, and that competitor figures come from public reports.
What we did not verify
We did not open an account, call gpt-6.1-sol, or measure a delay or a real price. “Available starting today” is the page’s sentence. The 29 September date comes from the product list, not from a timestamp repeated in the body of the announcement.
We did not read the safety addendum as a second source for this note. A safety figure quoted here would be a copy. The decision does not need that figure: the model is not local, and it is described as able to act on applications.
The decision
If you stay with the public chat, wait. The announcement does not reach you yet. If a work tool or an OpenAI API already answers for you, ask which model is connected, and what it is allowed to open.
Look elsewhere if you want the question to stay on the machine. This note does not say GPT-6.1 Sol is the best Sol. It says where it is, where it is not, and who receives the text.
Check and explore
Sources for this article
Numbers connect each reference to the passages that use it. Dates show when the documentation was consulted.
This article draws on the sources above. The exercises are for you to try on your devices; SecuFocus does not present them as tests carried out by its editorial team. Interfaces and features can change. Method and corrections.
Cite this article
Keep this reference with the article when you save or share it.