Published September 30, 2026
GPT 6.1 Sol replaced GPT 6 Sol everywhere the older model ran on Secret Chat AI today. OpenAI introduced it at DevDay yesterday, one week after GPT 6 Sol. There is nothing to install or select. Existing chats continue, and under each new reply the app shows the model that wrote it: “GPT 6.1 Sol.”
The practical change is in answers that already called for Sol. Web Search and Research now use the new version, as do the harder and web lookup paths through Smart Agentic ChatGPT. A quick answer still stays on GPT 6 Luna. OpenAI describes GPT 6.1 Sol as nearly matching GPT 6 Astra on agentic coding, computer use and professional work. That is OpenAI’s assessment, not a benchmark we ran.
What OpenAI Changed
OpenAI says GPT 6.1 Sol makes fewer factual errors than GPT 6 Sol and is more reliable at respecting explicit restrictions and the user’s intent in agentic tasks. Those are useful improvements for a model that reads web results and answers questions with several constraints. They are not a guarantee that any particular answer will be correct; check important claims against their sources.
The API model is gpt-6.1-sol. Its model page lists a 1,050,000-token context window, up to 128,000 output tokens, text and image input, and an April 30, 2026 knowledge cutoff. It supports low, medium, high, xhigh and max reasoning effort. Medium is OpenAI’s default. Secret Chat sends no reasoning-effort setting for this model, so that default applies here. The published context and output limits describe the API model; they do not promise that every Secret Chat request will use those limits.
What OpenAI’s Results Show
On DeepSWE v1.1, OpenAI says GPT 6.1 Sol matches GPT 6 Astra and beats GPT 6 Sol’s best score by 6.4 percentage points, at a lower reasoning effort than that Sol result. On AutomationBench 1.0.6, its medium-effort result is 4.8 percentage points above GPT 6 Sol at the same setting. Both are OpenAI’s published comparisons of agent tasks, not tests of Secret Chat’s chat modes.
OpenAI also reports that on the offline set of OSWorld 2.0, GPT 6.1 Sol leads GPT 6 Sol by seven percentage points at maximum reasoning effort and sits 2.1 points below GPT 6 Astra at maximum effort. OSWorld measures computer-use tasks; Secret Chat does not offer computer use through this model. These scores explain OpenAI’s capability claim, not a feature in our app.
Its factuality evaluation uses difficult conversations in which users had flagged an earlier model’s factual error. At low reasoning effort, OpenAI reports that answers containing an error fell from 11.4% for GPT 6 Sol to 7.7% for GPT 6.1 Sol. OpenAI says this set is deliberately hard and unrepresentative of typical use. Secret Chat uses the model’s medium default, so the low-effort figures should not be read as our expected error rate. All figures here belong to OpenAI’s evaluations.
Where GPT 6.1 Sol Runs on Secret Chat
OpenAI has no GPT 6 Terra middle model. As GPT 6 Sol did before, one Sol model covers both of our web-enabled ChatGPT modes:
- ChatGPT Web Search uses GPT 6.1 Sol with OpenAI’s built-in web search tool.
- ChatGPT Research uses GPT 6.1 Sol with the same built-in web search tool.
- Smart Agentic ChatGPT starts with GPT 6 Luna routing the question. A quick answer stays on Luna. A web lookup, harder question that needs thinking, or research goes to GPT 6.1 Sol.
- AI Council gives the ChatGPT seat to GPT 6 Luna on the lighter rung and GPT 6.1 Sol with web search on the heavier rung. GPT 6 Luna is the referee.
Free ChatGPT, without purchased credits, answers on GPT 6 Luna only. Web Search, Research and the harder Smart Agentic tiers need purchased credits, so free users do not reach GPT 6.1 Sol. GPT 6 Astra remains a separate manual mode; Smart Agentic never escalates to it. GPT 6 Luna also still handles Fast Chat, routing, chat titles and summaries.
What Changes in Generators
Email, Translator and Rewriter each offer a “ChatGPT Sol” row among six model choices. That row now runs GPT 6.1 Sol. The ChatGPT row in the Logo generator also runs GPT 6.1 Sol.
The agentic generators take a different path. In Article, Social Post, Document, Resume, Table, Chart and Presentation, the ChatGPT row answers on GPT 6 Luna, uses GPT 6.1 Sol for web lookups, and moves to GPT 6.1 Sol when the router escalates. That means choosing a ChatGPT row does not by itself mean every answer comes from Sol. Check the model label under the reply to see which model wrote it.
How It Fits Beside Luna and Astra
GPT 6.1 Sol costs no more than GPT 6 Sol did. It is much cheaper to run than GPT 6 Astra; GPT 6 Luna is cheaper still. That is why the routine quick answer stays on Luna while Astra remains a decision you make manually. If you need current sources, start with Web Search or Research. If you know a problem needs OpenAI’s top model, choose Astra yourself.
Answers written by GPT 6 Sol before the switch are still found and deleted the same way where the provider allows it, and the Session Privacy report lists them.
How Anonymous Access Works
Secret Chat sends your prompt to OpenAI without a personal OpenAI account or a profile built around you. We store no archive of chats on our servers, and prompts are never used for training.
The prompt goes to OpenAI verbatim so GPT 6.1 Sol can answer. Anonymity does not remove names or other identifying details from what you type. Redact sensitive details before sending; that part is your responsibility.
Frequently Asked Questions
- When did GPT 6.1 Sol arrive?
OpenAI introduced it at DevDay on September 29, 2026. It replaced GPT 6 Sol on Secret Chat AI on September 30, 2026.
- Do I need to start a new chat or select the new version?
No. Existing chats continue. In places that used GPT 6 Sol, new replies now come from GPT 6.1 Sol, and the model label under each reply says which model wrote it.
- Does ChatGPT Research use a different Sol model from Web Search?
No. OpenAI has no GPT 6 Terra middle model. Both Web Search and Research use GPT 6.1 Sol with OpenAI’s built-in web search tool.
- Will Smart Agentic ChatGPT always use GPT 6.1 Sol?
No. GPT 6 Luna routes each question and answers quick ones. Web lookups, harder questions that need thinking, and research go to GPT 6.1 Sol. Smart Agentic never escalates to GPT 6 Astra.
- Can a free ChatGPT answer use GPT 6.1 Sol?
No. Without purchased credits, ChatGPT answers on GPT 6 Luna. Web Search, Research and the harder tiers need purchased credits.
- What did OpenAI measure against GPT 6 Sol?
OpenAI reports a 6.4-point lead over GPT 6 Sol’s best score on DeepSWE v1.1 and a 4.8-point lead at medium effort on AutomationBench 1.0.6. Its low-effort factuality test found errors in 7.7% of GPT 6.1 Sol answers versus 11.4% for GPT 6 Sol. These are OpenAI’s results under its test conditions, not Secret Chat benchmarks.
- Which generators use the new version?
The “ChatGPT Sol” choice in Email, Translator and Rewriter and the ChatGPT Logo row run GPT 6.1 Sol. The ChatGPT row in Article, Social Post, Document, Resume, Table, Chart and Presentation uses Luna for answers and GPT 6.1 Sol for web lookups or router escalations.
- What does anonymous access protect?
You need no personal OpenAI account, Secret Chat builds no profile around you and stores no server chat archive, and prompts are never used for training. Your prompt still goes to OpenAI verbatim. Remove identifying details before sending.