When did GPT 6.1 Terra arrive?

OpenAI released it on October 26, 2026. It became ChatGPT’s middle tier on Secret Chat AI on October 27, 2026.

Do I need to select Terra or start a new chat?

No. Existing chats continue. The app routes eligible turns to GPT 6.1 Terra and shows “GPT 6.1 Terra” under a reply it wrote.

What is GPT 6.1 Terra best suited to?

OpenAI describes it as fast, with strengths in visual reasoning and clear, natural writing. Its examples include reading product photos or drawings and drafting or revising text.

Can it control my computer through Secret Chat AI?

No. OpenAI describes computer use as a model capability, but Secret Chat AI does not offer computer use through GPT 6.1 Terra.

Does ChatGPT Research now use Terra instead of Sol?

No. Manual Web Search uses GPT 6.1 Terra with web search. ChatGPT Research remains on GPT 6.1 Sol with web search, as do Smart Agentic’s research and deeper-thinking paths.

Can a free ChatGPT answer use GPT 6.1 Terra?

No. Free ChatGPT answers use GPT 6 Luna. Terra-backed Web Search, harder Smart Agentic routes, paid council panels and generators need a paid balance.

Can Terra read an image I attach?

Yes, when your turn runs on GPT 6.1 Terra. The picture reaches the model with your question. The ChatGPT family also accepts PDFs and text documents.

Does anonymous access remove names from my prompt?

No. Secret Chat AI sends the prompt to OpenAI verbatim. It stores no server chat archive, requires no personal OpenAI account, builds no profile around you, and never uses prompts for training. Remove identifying details before sending.

GPT 6.1 Terra Is Now on Secret Chat AI

Published October 27, 2026

GPT 6.1 Terra is now the middle tier of ChatGPT on Secret Chat AI. OpenAI released it yesterday. You will meet it in manual ChatGPT Web Search, the web lookup path of Smart Agentic ChatGPT, the ChatGPT seat of the AI Council when a question turns on current or checkable facts, and the ChatGPT Terra row in several generators. There is nothing to install or select for existing chats. The model name under each reply tells you when Terra wrote the answer.

Since September 23, the middle tier had borrowed the flagship model: first GPT 6 Sol, then GPT 6.1 Sol. It has its own model again. That changes which model handles a web lookup, but it does not move ChatGPT Research or harder reasoning away from GPT 6.1 Sol.

What OpenAI Built GPT 6.1 Terra For

OpenAI describes GPT 6.1 Terra as “a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing.” Its examples are concrete: match product photos to catalog entries, interpret or compare technical drawings, act on application screens, and draft or revise text. OpenAI also describes it as suited to high-volume professional work, with strong image analysis and computer use capabilities. These are OpenAI’s descriptions of the model, not results measured on Secret Chat AI.

On Secret Chat AI, the relevant uses are reading an image you attach and working with text. The ChatGPT family accepts pictures, PDFs and text documents. When a turn runs on GPT 6.1 Terra, an attached picture reaches the model with your question. That lets you ask about a product photo, screenshot or drawing in a Terra-backed mode. OpenAI’s “clear, natural writing” claim is most relevant to the ChatGPT Terra rows in Email, Translator and Rewriter, which always write on GPT 6.1 Terra. You can also revise a draft in chat. Secret Chat AI does not offer computer use through this model: it cannot act on application screens for you here.

What Stands Out in OpenAI’s Description

Speed is part of OpenAI’s stated purpose for Terra. For factuality, its system card reports a more specific result. On de-identified ChatGPT conversations that users had flagged for factual errors, OpenAI found GPT 6.1 Terra made substantially fewer factual errors than GPT 6 Sol. It was also less likely to repeat the particular error a user had reported. OpenAI says its performance on that test was on par with GPT 6.1 Sol.

That test deliberately used difficult, error-prone conversations. OpenAI warns that its absolute error rates should not be read as rates in production. Its evaluations ran in OpenAI’s research environment or through its API, not on Secret Chat AI. It is OpenAI’s result, not ours, and an answer can still be wrong. Check important claims against their sources.

The API model page lists a June 1, 2026 knowledge cutoff, about a month later than GPT 6.1 Sol’s April 30 cutoff. A later cutoff gives the model later training knowledge; it is not a substitute for a current web lookup. OpenAI lists a 1,050,000-token context window and a 128,000-token maximum output, the same published limits as GPT 6.1 Sol. Those are API limits, not a promise that every request in Secret Chat AI will use them.

Terra accepts text and image input and produces text. OpenAI lists web search, file search, streaming and structured outputs among its supported capabilities. The API identifier is gpt-6.1-terra. OpenAI offers reasoning effort settings from none through max, with medium as the default. Secret Chat AI sends no reasoning-effort setting for GPT models, so that default applies. This is the first Terra model in the GPT-6 generation, following GPT 5.6 Terra.

Where It Runs in Secret Chat AI

The clearest place to use Terra is ChatGPT Web Search. It is a manual mode and runs GPT 6.1 Terra with OpenAI’s built-in web search tool. Use it when the answer depends on something current or checkable. The model’s training cutoff and a live lookup serve different purposes.

Smart Agentic ChatGPT makes the choice for each question. GPT 6 Luna reads it first. A quick answer stays on Luna; a question needing a web lookup goes to GPT 6.1 Terra with web search. A harder question that needs deeper thinking, or research, goes to GPT 6.1 Sol. Smart Agentic never escalates to GPT 6 Astra.

In the AI Council, the ChatGPT seat uses GPT 6.1 Terra with web search when the panel is set to the stronger rung for current or checkable facts. The lighter rung still uses GPT 6 Luna. Luna remains the referee. The free council pair, DeepSeek and GPT Luna, is unchanged.

Which Generators Use Terra

Email, Translator and Rewriter each show seven fixed model rows: ChatGPT Luna, Claude Haiku, Gemini Flash, Grok 4.7, ChatGPT Terra, Claude Sonnet and GLM Abliterated. The ChatGPT Terra row now runs GPT 6.1 Terra. While the middle tier borrowed Sol, that row was labelled ChatGPT Sol.

The ChatGPT row in the Logo generator now draws its logos on GPT 6.1 Terra by default. Its router can still move an especially hard request to GPT 6.1 Sol. Article, Social Post, Document, Resume, Table, Chart and Presentation use a different path: their ChatGPT row answers on GPT 6 Luna, uses GPT 6.1 Terra for web lookups, and moves hard requests to GPT 6.1 Sol when the router escalates. Article’s deep-research option also runs on GPT 6.1 Sol.

So a ChatGPT row is not a promise that Terra wrote every answer. Look under the reply for the model label. A Terra answer says “GPT 6.1 Terra.” Existing chats continue, and the label reflects the model that wrote each new reply.

What Stays on Sol, Astra and Luna

ChatGPT Research stays on GPT 6.1 Sol with web search. So does Smart Agentic’s deeper-thinking path. GPT 6 Astra remains a separate manual mode. GPT 6 Luna still handles Fast Chat, routing, chat titles and summaries, as well as free ChatGPT answers.

Without a paid balance, free ChatGPT answers use GPT 6 Luna only. Web Search, Research, the harder Smart Agentic tiers, paid council panels and generators need a paid balance. Free users therefore do not reach GPT 6.1 Terra. Answers written by GPT 6.1 Sol in the Web Search tier before this 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 AI requires no personal OpenAI account and builds no profile around you. It stores no archive of chats on its servers, and prompts are never used for training. These are the privacy promises of the service.

Your prompt goes to OpenAI verbatim so the model can answer. Anonymity does not remove names or identifying details you type or attach. Redact sensitive details before sending; that is your responsibility.

Frequently Asked Questions

When did GPT 6.1 Terra arrive?
OpenAI released it on October 26, 2026. It became ChatGPT’s middle tier on Secret Chat AI on October 27, 2026.
Do I need to select Terra or start a new chat?
No. Existing chats continue. The app routes eligible turns to GPT 6.1 Terra and shows “GPT 6.1 Terra” under a reply it wrote.
What is GPT 6.1 Terra best suited to?
OpenAI describes it as fast, with strengths in visual reasoning and clear, natural writing. Its examples include reading product photos or drawings and drafting or revising text.
Can it control my computer through Secret Chat AI?
No. OpenAI describes computer use as a model capability, but Secret Chat AI does not offer computer use through GPT 6.1 Terra.
Does ChatGPT Research now use Terra instead of Sol?
No. Manual Web Search uses GPT 6.1 Terra with web search. ChatGPT Research remains on GPT 6.1 Sol with web search, as do Smart Agentic’s research and deeper-thinking paths.
Can a free ChatGPT answer use GPT 6.1 Terra?
No. Free ChatGPT answers use GPT 6 Luna. Terra-backed Web Search, harder Smart Agentic routes, paid council panels and generators need a paid balance.
Can Terra read an image I attach?
Yes, when your turn runs on GPT 6.1 Terra. The picture reaches the model with your question. The ChatGPT family also accepts PDFs and text documents.
Does anonymous access remove names from my prompt?
No. Secret Chat AI sends the prompt to OpenAI verbatim. It stores no server chat archive, requires no personal OpenAI account, builds no profile around you, and never uses prompts for training. Remove identifying details before sending.

GPT 6.1 Terra Is Here:
ChatGPT's Middle Tier Gets Its Own Model Again