On September 29, 2026, OpenAI announced GPT-6.1 Sol for ChatGPT Work and Codex. It said the model improves on GPT-6 Sol at coding, using computers, and work tasks. If you mainly use ChatGPT for questions or emails, one detail matters most: GPT-6.1 Sol was not yet available in regular Chat at launch.
OpenAI also shared test results and API prices lower than GPT-6 Astra’s. Keep the source in mind: these are OpenAI’s tests, and most cover tasks unlike an everyday chat.
What Happened
GPT-6.1 Sol is an update to GPT-6 Sol. OpenAI aims it at complex coding, using software, and tasks with several steps. Think of asking AI to finish a job, rather than answer one quick question.
At the September 29 announcement, OpenAI listed access for Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. Availability depends on your plan, client and workspace settings, as OpenAI’s current documentation explains. That plan list does not guarantee access in every client or workspace. OpenAI said it wasn’t yet in regular Chat and gave no date for that to change.
That matters if you’re eyeing a Plus or Pro plan just for this model. Check where you plan to use it. Access in ChatGPT Work or Codex doesn’t mean you’ll find Sol in your usual chat window.
How GPT-6.1 Sol Stacks Up
OpenAI reports gains over GPT-6 Sol in two tests it highlighted. It also says Sol gets close to the pricier GPT-6 Astra in a test where AI uses a computer. Here are the numbers with their test settings attached (a percentage needs context, like a speed limit needs a road sign).
| Test and what it measures | OpenAI-reported comparison | What it means for readers |
|---|---|---|
| DeepSWE v1.1: Complex coding tasks | GPT-6.1 Sol exceeded GPT-6 Sol’s best result by 6.4 percentage points at lower reasoning effort and cost, according to OpenAI. | Promising for hard coding work. It says little about how a normal ChatGPT conversation will feel. |
| OSWorld 2.0: Completing tasks on a computer | GPT-6.1 Sol gained seven percentage points over GPT-6 Sol at maximum reasoning effort and came within 2.1 points of GPT-6 Astra, according to OpenAI. | Useful if you ask AI to work in apps. It doesn’t mean every task will go right. |
| Factual-error test: Whether an answer has a factual mistake | At low reasoning effort, the share of answers with a factual error fell from 11.4% to 7.7%, according to OpenAI. | That’s a welcome drop, but check answers that matter. |
“Reasoning effort” is how much work the model can spend on an answer. The settings matter. The factual-error result used low effort; the computer-use comparison used maximum effort. Neither score tells you how Sol handles every question.
What about Claude and Google?
OpenAI’s announcement compares Sol with Claude Opus 5.5. It reports that Sol scored 2.2 points higher on AutomationBench 1.0.6 at medium reasoning effort and cost roughly one-third as much per task in that test. Those figures come from OpenAI. They’re useful for that test, but they don’t tell us which model writes a better email or handles every work task.
A ranking against Google’s current model would need matched tests under the same conditions. The same goes for a broader comparison with Claude. If you use these tools for work, try them on your tasks. A test score can’t tell you how a model will handle your spreadsheet or your company’s software.
Does it cost more?
For developers and businesses using the API, OpenAI lists GPT-6.1 Sol at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. A token is a small piece of text. The API bill depends on how much text goes in and comes out. The model documentation lists higher rates for requests above 272,000 input tokens and for Fast mode. OpenAI says Sol gets close to GPT-6 Astra’s performance at about one-fifth of Astra’s standard token price. Those are OpenAI’s published prices and comparison; the cost of a task still depends on how you use the model.
Those are API rates, not ChatGPT subscription prices. Your Plus or Pro bill doesn’t become $2 or $10. The announcement doesn’t set a new subscription price for Sol in ChatGPT Work or Codex. If you only use the ChatGPT app or website, those token rates are mostly background.
The low cached-input rate may help teams that send the same instructions again and again. It’s like paying less for text the system has already seen, when it meets the right conditions. If you ask ChatGPT a few questions each day, it probably won’t change what you pay.
Our Take
Verdict: GPT-6.1 Sol is worth a look if you use Codex or ChatGPT Work for coding and tasks with several steps. OpenAI’s reported gains give you a reason to try it on work you already do. If your team uses the API, compare the token rates with your own usage. A lower price per task depends on the task.
For everyday ChatGPT users, there’s no rush to change plans. Sol wasn’t available in regular Chat at launch, and its headline tests focus on specialized work. Check OpenAI’s announcement for access details before paying for a plan just to use Sol.