Oct 27, 2026

How to Use GPT-6.1 Terra in ChatGPT Work and Codex: Where to Find It, Which Plans Get It, When to Pick It

If you pay for ChatGPT Plus, Pro, Business, Enterprise or Edu, you can switch to GPT-6.1 Terra today in ChatGPT Work and Codex: open the model control under the message box, choose Advanced, and pick GPT-6.1 Terra from the model list. It is OpenAI’s new fast, cheaper model for writing, reading screenshots and images, and operating apps and websites for you. The practical reason to use it is usage: OpenAI’s own estimate for Plus is about 35 to 285 messages per five hours with Terra, versus 15 to 160 with GPT-6.1 Sol and 5 to 45 with GPT-6 Astra. It is not in the regular Chat tab, and for the hardest coding work OpenAI still recommends Sol or Astra. Everything below was checked against OpenAI’s announcement, release notes and ChatGPT Learn docs on October 27, 2026.

Key facts (from OpenAI, October 26–27, 2026):

Where: ChatGPT Work and Codex only (desktop app, Work on the web, Codex CLI and IDE extension). Not in regular Chat. (release notes)

Plans: Plus, Pro, Business, Enterprise and Edu. Not Free or Go. It is rolling out, so it may reach your account later in the day.

Usage: your plan’s overall limits are unchanged, but Terra uses them more slowly: 35–285 local messages per five hours on Plus and Standard Business (pricing).

Cost on credits: 25 credits per 1M input tokens and 125 per 1M output, half of GPT-6.1 Sol’s 50/250.

Codex CLI: needs version 0.159.0 or later; start it with codex -m gpt-6.1-terra (models docs).

Default may change: some Plus and Pro users will see GPT-6.1 Terra at Medium instead of GPT-6.1 Sol at Light in Work’s default Power presets.

If you have not used the GPT-6 lineup yet, our GPT-6.1 Sol switching guide explains the Astra / Sol / Luna names that Terra now joins. More guides like this one are on the 7minai homepage.

What GPT-6.1 Terra is, in plain terms

OpenAI’s launch post describes Terra as “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing.” It replaces GPT-5.6 Terra as the middle “fast and efficient” option, and OpenAI calls it “a significant step up” from that model.

In everyday terms, it is built for the kind of work that fills an office day:

Writing: drafting proposals, emails, summaries and reports in a natural tone.
Understanding images and screens: reading dashboard screenshots, charts, technical drawings, product photos and scanned forms.
Doing things on a computer for you: filling in forms from source documents, entering data into apps, clicking through websites and desktop apps when Work or Codex has permission to use them.

OpenAI’s ChatGPT Learn models page sums up the four current models like this:

Model	OpenAI’s one-line positioning	Best for
GPT-6 Astra	”The hardest end-to-end work”	Complex research, code and decisions that need sustained judgment
GPT-6.1 Sol	”Repeated, long-running work” when cost matters	Long coding sessions, multi-source documents, agent work
GPT-6.1 Terra	”Everyday writing and visual work”	Drafting, reading images and screens, filling forms, app tasks
GPT-6 Luna	”Clear, repeatable tasks”	High-volume extraction, classification, structured summaries

What the benchmarks say (and what they don’t)

These numbers come from OpenAI’s own launch post, so treat them as the vendor’s claims rather than independent results:

Computer use: on Agents’ Last Exam, a set of professional tasks such as turning architectural drawings into a 3D model, Terra at medium effort scored 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, and OpenAI says it “performs on par with GPT-6 Astra at only one-tenth its price per token.”
Speed: OpenAI says Terra completes tasks “over three times faster than both GPT-6 Astra and GPT-6.1 Sol.”
Fewer made-up facts: on OpenAI’s internal set of error-inducing prompts, the error rate at low effort fell from 53.6% (GPT-5.6 Terra) to 16.7%, and to 4.3% at maximum effort.
Coding: on Terminal-Bench 4.0 at maximum effort, Terra solved 50.3% of tasks versus 21.5% for GPT-5.6 Terra. But OpenAI is explicit: “For the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice.”

Artificial Analysis, an independent benchmarking site, published its own evaluation saying Terra is “Pareto-optimal” (the best intelligence for the money) at low and medium effort. At max effort it scored 4 points below GPT-6.1 Sol on their Intelligence Index, and 48 on their Coding Agent Index versus 54 for Sol and 31 for Luna. The honest takeaway: Terra is a better value model, not a smarter-than-Sol model.

Who can use it

Your plan	GPT-6.1 Terra in Work and Codex?	Notes
Free, Go	No	These plans get GPT-6 Luna in the desktop app
Plus ($20/month)	Yes, rolling out	Same overall limits; Terra stretches them further
Pro ($100 / $200 / $500)	Yes, rolling out	Pro plans currently have no five-hour limit
Business	Yes, rolling out	Same per-message estimates as Plus on Standard Business
Enterprise, Edu	Yes, rolling out	Depends on workspace availability and admin controls

Two extra limits from OpenAI’s pricing page and release notes:

It is not in Chat. The regular Chat tab uses GPT-6 Sol (paid) or GPT-6 Luna (Free and Go). Terra, like GPT-6.1 Sol, lives only in Work and Codex.
No UAE inference residency. If your Enterprise workspace requires processing in the UAE, Terra is not available. OpenAI notes this depends on where inference runs, not where you are.

How to switch to Terra in the ChatGPT desktop app and Work on the web

You do not need any code or settings file for this. In Work or Codex inside the ChatGPT desktop app, or in ChatGPT Work on the web:

Open Work (or Codex) rather than Chat. In the desktop app the switcher at the top lets you choose ChatGPT or Codex; inside ChatGPT, pick Work. On the web, open ChatGPT Work.
Find the model and reasoning control below the message box. By default it shows a “Power” slider of presets. OpenAI’s docs show examples such as Luna High, GPT-6 Sol Light, GPT-6 Sol Medium, Astra Light, Astra Medium and Astra Extra High; the exact presets you see depend on your plan and workspace.
Open Advanced and choose GPT-6.1 Terra from the model list. OpenAI’s instruction is literally: “To try GPT-6.1 Terra … open Advanced and choose it from the model list.”
Pick a reasoning effort. Start at the default (Medium for most people). Use Light for quick rewrites and short emails; move to High or Extra High only for multi-step tasks with several sources.
Leave speed on Standard unless you need it fast. Terra supports Standard and Fast where available. Fast mode uses your included allowance 2.5x faster (and 2x the credit rate on purchased credits), which cancels much of Terra’s usage advantage.

The model you pick manually sticks. OpenAI’s Work and Codex help article says Codex “keeps the model you selected manually,” so you will not be silently switched back.

Heads-up about the default. OpenAI’s release note says “some Plus and Pro users may see GPT-6.1 Terra at Medium instead of GPT-6.1 Sol at Light in the default Power presets in Work.” If your Work results suddenly feel different and you never changed anything, check the control under the message box: you may already be on Terra. If you preferred the old behavior, select GPT-6.1 Sol under Advanced.

How to use Terra in Codex

Codex in the desktop app or IDE extension: use the same model switcher under the composer and pick GPT-6.1 Terra.

Codex CLI (the terminal version): this is the one place you need a command, and it is short.

Update the CLI first. Terra requires Codex CLI 0.159.0 or later. If you installed Codex through your AI agent or a package manager, simply ask it to “update the Codex CLI to the latest version.”
Start a session with Terra: codex -m gpt-6.1-terra
Inside a running session, type /model to switch models or change reasoning effort.
To make Terra your default everywhere (desktop app, CLI and IDE extension share one settings file, config.toml), add the line model = "gpt-6.1-terra". The easiest no-code route is to tell Codex: “Set my default model in config.toml to gpt-6.1-terra.”

Should you make it your Codex default? For everyday edits, explanations, documentation and UI tweaks from screenshots, Terra is a reasonable default because it stretches your limits. For large refactors, hard bugs or long autonomous runs, OpenAI’s guidance is to stay on GPT-6.1 Sol or Astra. A sensible split many people will land on: Terra by default, switch with /model (or the picker) when a task gets hard.

Before you hand a cheaper model more autonomy in a real repository, keep the same review habits as before. Our Codex Security guide covers safe permissions and reviewing patches.

How much usage you actually get

OpenAI does not publish fixed message caps for Work and Codex. Its pricing page gives ranges of local messages per five-hour period for Plus and Standard Business:

Model	Plus	Standard Business
GPT-6 Astra	5–45	5–45
GPT-6.1 Sol	15–160	15–160
GPT-6.1 Terra	35–285	35–285
GPT-6 Sol	15–150	15–150
GPT-6 Luna	350–3,000	350–3,000

What to know about these numbers:

They are estimates, not limits. Long context, tool use, big files and high reasoning effort all use more per message. Cloud tasks may use more than local ones. Weekly limits may also apply.
Work and Codex share one allowance. Messages in ChatGPT Work and in Codex come out of the same budget, and so does ChatGPT for Excel on Plus and Pro.
Pro plans currently have no five-hour limit, so for Pro users the benefit is mostly speed and a slower drain on weekly or credit-based usage.
Purchased credits: Terra costs 25 credits per 1M input tokens, 1.25 per 1M cached input and 125 per 1M output. That is half of GPT-6.1 Sol (50 / 2.5 / 250) and one-tenth of GPT-6 Astra (250 / 25 / 1,250).
Check your real remaining usage in the usage dashboard, or type /status in a Codex CLI session.

If you are weighing whether a Pro tier is worth it, our note on the Pro 200 usage change and the new Pro 500 tier explains how the Pro plans now differ.

What to give Terra (and what to keep on Sol or Astra)

OpenAI’s models page includes example prompts for each model. The Terra example is a good template for the kind of task it is built for:

“Review these dashboard screenshots and summarize the main changes. Refer to the chart labels and values, and flag anything you can’t read confidently.”

Good fits for Terra, based on OpenAI’s own descriptions:

Screenshots and charts: “Summarize what changed between these two dashboard screenshots” or “Read this chart and list the three biggest movers.”
Drafting: proposals, client emails, meeting follow-ups, product descriptions.
Forms and data entry: “Fill out this vendor form using the details in the attached PDF,” or entering invoice details into an app (OpenAI’s own API example is reading invoices and entering them into accounting software).
Technical drawings and photos: interpreting a floor plan or matching product photos to catalog entries.
Website and app tasks: in Work, browsing and completing steps on sites you allow. If you want an agent to work inside your own signed-in browser, see our guide to letting AI agents use your logged-in Chrome.

Keep these on a bigger model:

Complex coding across many files, hard debugging and architecture: GPT-6.1 Sol or Astra.
High-stakes decisions that weigh many sources and tradeoffs: Astra, which OpenAI says is better at asking focused questions while keeping your constraints in view.
Massive repetitive jobs where you know exactly what good output looks like (tagging, extraction): Luna is far cheaper still.

Always check details taken from images. OpenAI’s own advice on the models page is to “review the result against your source materials, especially names, numbers, and details taken from images.” Terra’s prompt example deliberately asks it to flag anything it can’t read confidently; copy that phrase into your own prompts.

Troubleshooting

I don’t see GPT-6.1 Terra anywhere. Work through these in order:

Are you in Work or Codex? It is not in the regular Chat tab.
Are you on Plus, Pro, Business, Enterprise or Edu? Free and Go are not included.
Did you open Advanced in the model control? Terra may not be one of the default Power presets.
It is a gradual rollout. OpenAI says it becomes available “as it reaches your account.” Update the desktop app and check again later.
On Enterprise, access depends on workspace availability and admin controls; ask your ChatGPT admin. Workspaces with UAE inference residency can’t use it.

Codex CLI says the model doesn’t exist or isn’t supported. Your CLI is probably older than 0.159.0. Update it, then run codex -m gpt-6.1-terra again. Check the spelling: the ID is gpt-6.1-terra with a dot, not a dash, between 6 and 1.

My Work results changed and I didn’t touch anything. Some Plus and Pro users had their default preset changed to GPT-6.1 Terra at Medium. Check the control under the composer and pick GPT-6.1 Sol under Advanced if you want the previous default back.

My usage is draining faster than the table suggests. Check whether Fast mode is on (2.5x included usage), whether you are attaching large files or long chat histories, and how many plugins or MCP servers are connected. OpenAI’s tips: keep prompts precise, give only the relevant files, define the output length, and disable MCP servers you don’t need.

Terra misread a number in a screenshot. Ask it to quote the exact label and value it read, raise the reasoning effort for that task, or send a sharper crop of the image. For anything that goes into a report or a form submission, verify against the source.

FAQ

Is GPT-6.1 Terra better than GPT-6.1 Sol? Not overall. It is faster and about half the cost per token, and OpenAI says it surpasses Sol on one science-coding benchmark at max effort, but OpenAI still recommends Sol and Astra for the most complex coding. Think of Terra as the better-value everyday model.

Is it free? It is included in paid plans at no extra charge (Plus, Pro, Business, Enterprise, Edu), drawing from your existing Work and Codex allowance. Free and Go plans don’t get it.

Can I use it in the normal ChatGPT chat? No. As of October 27, 2026, OpenAI lists it only for Work and Codex.

Does Terra replace GPT-5.6 Terra? OpenAI positions it as the successor and “a significant step up,” and the pricing page says GPT-5.6 Terra’s credit rates (50 / 5 / 300) are unchanged, so GPT-5.6 Terra still has a listed rate for now. GPT-6.1 Terra is cheaper per token than GPT-5.6 Terra on credits.

Is it safe to let it operate my computer? OpenAI says Terra meets its “Critical” cybersecurity threshold and keeps the protections built for GPT-6 Astra, and that it showed strong resistance in prompt-injection tests. That is not a guarantee: keep confirmations on for purchases, submissions and messages, and only grant access to sites and apps the task needs.

Optional appendix: the API details

Most readers can skip this. If you or a tool you use calls the OpenAI API directly, the model ID is gpt-6.1-terra (model page):

Price: $1 per 1M input tokens, $0.05 cached input, $1.25 cache writes (30-minute retention), $5 output. Batch and Flex are 50% cheaper; Fast mode is 2x Standard. Prompts over 272K input tokens are billed at 2x input and 1.5x output for the whole request.
Limits: 1,050,000-token context window (922,000 max input), 128,000 max output, knowledge cutoff June 1, 2026. Text and image in, text out.
Reasoning effort: none, low, medium (default), high, xhigh, max. Unlike GPT-6.1 Sol, Terra accepts none.
Tools (web search, file search, computer use, MCP and others) require the Responses API; Chat Completions works for plain requests.

If you are moving existing code from a GPT-5.x model, the four migration changes in our GPT-6.1 Sol guide apply here too, except the none effort restriction.