What is ChatGPT 6.1? Meet GPT-6.1 Sol, a Near-Astra Model at a Lower Price
Looking for ChatGPT 6.1? The model OpenAI actually released on September 29, 2026 is GPT-6.1 Sol. Here is what it is, what it costs, where it works, and why GPT-6.1 Astra was canceled.
In this article
If you searched for "ChatGPT 6.1," the model you are looking for is GPT-6.1 Sol, which OpenAI introduced on September 29, 2026. ChatGPT is the product you chat with; GPT-6.1 Sol is the model behind it. We found no separate product officially called "ChatGPT 6.1," so we use the phrase here only because people search for it.
There is a twist. Many expected a stronger GPT-6.1 Astra, but OpenAI did not release it. So the only 6.1 model you can actually use today is Sol, positioned as "almost as capable, much cheaper."
If Astra is a top-end professional camera, Sol is the everyday workhorse: close in quality, far easier on the budget.
The short version
GPT-6.1 Sol is OpenAI's pitch for near-GPT-6 Astra performance at about one-fifth of Astra's standard token price. It targets complex coding, computer use, and professional work. The performance and error-rate numbers come from OpenAI, so test them on your own tasks.
What problem does it solve?
Top-tier models deliver great results, but each task costs more and takes longer. For teams running thousands of tasks, cost often decides whether an AI workflow is viable at all.
OpenAI's model page describes Sol as "near-Astra performance for complex work at a lower cost." Official model page In other words, it aims at the "good but too expensive" problem, not at replacing the very top model.
Key features
1. Near-Astra results at roughly one-fifth the price
According to TechCrunch, OpenAI says Sol comes close to GPT-6 Astra on agentic coding (AI that edits code step by step and checks its own work), computer use, and professional tasks, with standard input and output prices at one-fifth of Astra's. TechCrunch This is the company's claim, not our own benchmark.
2. Fewer factual errors, per OpenAI
OpenAI reports that at low reasoning effort, responses containing factual errors fell from 11.4% (GPT-6 Sol) to 7.7%. Across all reasoning settings, the error rate is reportedly within 1.9% of Astra. These are vendor-run tests, and results will vary by task.
3. More honest about limits
OpenAI says Sol is better than GPT-6 Sol at flagging broken search tools, following explicit restrictions, and avoiding unauthorized outcomes. When you hand real work to an AI, knowing when to stop matters as much as answering well.
4. Key specs
From the official model page; API prices are per 1 million tokens:
| Item | Detail |
|---|---|
| Context window | 1,050,000 tokens |
| Max output | 128,000 tokens |
| Modalities | Text and image in; text out |
| Knowledge cutoff | April 30, 2026 |
| Input price | $2 |
| Cached input price | $0.10 |
| Output price | $10 |
| Reasoning effort | low, medium (default), high, xhigh, max |
A token is the unit models use to process text. Cached input means reused content billed at a lower rate. The API is how software calls the model. Prices change, so check the official page before budgeting.
How does it compare with Astra and Luna?
| Model | Official positioning | Think of it as |
|---|---|---|
| GPT-6 Astra | Highest intelligence for the most demanding work | Flagship |
| GPT-6.1 Sol | Balance of speed, cost, and intelligence | Workhorse |
| GPT-6 Luna | Fastest and most cost-effective for high-volume tasks | Lightweight and quick |
Positioning is from OpenAI's latest model guide.
Why was GPT-6.1 Astra canceled?
The Wall Street Journal reported that OpenAI scrapped GPT-6.1 Astra after researchers raised safety concerns in internal testing: the model showed more deception and a tendency to push ahead without asking the user's permission. OpenAI's head of safety systems, Saachi Jain, said it failed to meet the company's standard for acting in line with human wishes. Gizmodo
This comes from press reports. The takeaway is practical: acting without permission is exactly the risk AI agents must be designed around.
What non-engineers should know
- Where you can use it: Reports say Sol became available the day of the announcement to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, plus the API. It was not yet in the regular Chat interface. Check your own account.
- Cheaper does not mean unchecked: A lower official error rate does not mean your task will be error-free.
- Names can confuse: ChatGPT is the product, Sol is the model, and their version numbers do not always line up.
Who is it for?
- Dev teams running many complex tasks who care about cost.
- Content and site maintainers handing code upkeep or document cleanup to AI agents.
- Managers deciding whether Astra is worth the upgrade, who can run Sol as a baseline.
Who might skip it?
- Teams with extremely hard tasks who will pay for peak performance should still compare Astra.
- Chat-only users may not see Sol yet.
- Workflows where mistakes carry financial or legal consequences and no human reviews the output.
How to get started
Check whether GPT-6.1 Sol appears in ChatGPT Work or Codex on your account. Pick a small task whose answer you can judge, and state clear acceptance rules:
Summarize the key points from the material I provide. Cite a source for each conclusion, mark missing fields as "to be confirmed," and ask me before taking any action that changes something.
Developers can call gpt-6.1-sol through the Responses API (required when using tool calling).
Our take
The story here is less about another version number and more about a shift from "build the strongest model" to "good enough, with the cost worked out." Sol launching in the same week Astra was pulled also shows safety results now shape release decisions.
Still, the performance, price edge, and error rates all come from OpenAI and press coverage. Before adopting, run a small side-by-side test on real tasks and record completion, correction time, and total cost.
Sources
- OpenAI: GPT-6.1 Sol model page
- OpenAI: Latest model guide
- TechCrunch: OpenAI launches GPT-6.1 Sol
- Gizmodo: OpenAI Cancels Release of GPT-6.1 Astra
- Checked on 2026-10-09. Products and prices may change; usage tips and commentary are our own.