GPT-6.1 Terra: Features, Model Differences, Pricing and Setup

GPT-6.1 Terra is an OpenAI model focused on understanding images and screens, acting through supported tools, and drafting or revising clear text. It is worth considering for work that connects visual material with documents or applications, with separate access requirements for ChatGPT Work, Codex, and the API (application programming interface, a way for software to send requests to the model and receive results).

What is GPT-6.1 Terra?

The OpenAI API changelog dates the release to October 26, 2026. Its API identifier is gpt-6.1-terra. A model interprets inputs and produces answers or tool calls; the product and connected tools supply the environment that actually handles files, browsers, and applications.

A useful question is whether Terra can read the supplied material accurately and complete a task with a defined scope. Matching photographs to product records or shortening a proposal while preserving its evidence are practical candidates. Results still depend on input quality, available tools, and the criteria used to review the work.

Key features: visual reasoning, computer use, and writing

Visual reasoning means interpreting objects, layout, and relationships in an image to make a useful judgment. A chart requires attention to axes and units; an application screen requires understanding which item is selected and what a control affects. Provide an enlarged crop or separate text for tiny labels and ambiguous numbers. Asking the model to guess information it cannot see makes verification harder.

Computer use connects interpretation with action. Selecting Terra does not automatically give it access to every application: a supported browser or computer tool, the relevant login, and appropriate permissions are required. Reopen the saved result afterward. A success message and a correctly updated record are two different things to check.

For writing, provide source material, the intended audience, the purpose, and the information that must survive revision. “Shorten this customer notice while retaining prices, conditions, and exceptions” gives a clearer acceptance criterion than “Make it shorter.” These uses align with OpenAI's Terra availability announcement; they do not establish that Terra wins every comparison.

How does Terra compare with Sol, Astra, and Luna?

Choose by task difficulty and required outcome rather than model numbering or release order. This table organizes the official model selection guidance into a starting point for evaluation.

Terra is a reasonable comparison candidate for finding product matches in photographs and producing a list for human review. Sol or Astra also deserve consideration when a task changes several systems and requires decisions about exceptions and regression risks. For frequent, simple classification, first check whether Luna already meets the quality bar.

Use the same material, instructions, and reasoning effort when comparing. Record missing facts, incorrect changes, and review time alongside response time. A quick answer that takes longer to repair may not save time overall. The table describes selection criteria, not performance measurements made for this article.

Where can you use it?

The launch availability notice lists ChatGPT Work and Codex on Plus, Pro, Business, Enterprise, and Edu. Terra is unavailable in ordinary Chat. Check Advanced in the model list in the desktop app or Work on the web; Work mobile is also listed as supported.

For organizational accounts, workspace availability and administrator controls matter too. If the model is missing, check that you are using Work or Codex, then check the app version, account, and workspace. Browser use and computer use have separate setup and access requirements, so seeing the model does not establish that every action tool is available.

Organizations with processing-location requirements should also note that Terra does not support UAE inference residency. That condition concerns where inference runs; it should not be restated as a ban on all users located in the UAE.

How do pricing and usage work?

Included usage and credits in Work or Codex should be distinguished from token billing for developer API calls. Tokens are units used to process text and other inputs, not a one-to-one count of words or characters. Check subscription usage in your account and the plan documentation; check API costs in API usage records and its price table.

The following Terra Standard API text rates apply at its October 2026 launch. Each price is in US dollars per one million tokens. The official API pricing page distinguishes processing tiers and cache writes.

As a calculation example, 10,000 uncached input tokens and 2,000 billable output tokens cost US$0.01 + US$0.01 = US$0.02. This assumes Standard processing without cache writes, extra tools, regional premiums, or other additions. It is not a quote for a real workflow: use actual usage, including reasoning and repeated calls.

Above 272,000 input tokens, the whole request uses twice the input and cache rates and 1.5 times the output rate. Fast is priced at twice Standard; Batch and Flex are 50% below Standard. Regional processing carries a 10% premium where offered, and tools may add charges. Selecting relevant source sections before sending a large bundle can help both accuracy and cost control.

Getting started in the app, Codex CLI, and API

In the app: Open the model selector near the composer in Work or Codex, choose GPT-6.1 Terra, attach the necessary material, and specify the output and permitted changes. Start with an easy-to-review analysis or draft. The location of model and reasoning settings can differ between clients.

In Codex CLI: The command-line interface runs Codex from a terminal. The official model documentation requires Codex CLI 0.159.0 or later for Terra. With Codex installed and signed in, enter your working directory and run:

codex -m gpt-6.1-terra

Use /model in an interactive session to inspect or change the model and reasoning effort. If Terra is unavailable, check the installed version, account access, and exact identifier. Model selection does not replace permission settings for file edits or external changes.

Through the API: You need Python, an API project with usable billing and model access, and a project API key. Set the key in the execution environment as OPENAI_API_KEY; do not place it in code or a public repository. The official API quickstart provides operating-system-specific environment setup.

1. Install the SDK (software development kit, reusable code and tools that make the API easier to use) in the Python environment you will run:

   python -m pip install --upgrade openai

2. Save this complete example as terra_example.py. It checks a text request without connecting tools that browse or modify files.

   from openai import OpenAI
   
   client = OpenAI()
   response = client.responses.create(
       model="gpt-6.1-terra",
       reasoning={"effort": "low"},
       input=(
           "Turn these notes into a three-sentence customer notice in English. "
           "Preserve conditions and add no facts. "
           "Notes: Friday delivery. In-stock items only. Address changes by Thursday noon."
       ),
   )
   print(response.output_text)

3. Run the saved example file with this command:

   python terra_example.py

   The request should print a notice. Wording can vary; check that Friday delivery, the in-stock restriction, and Thursday noon's address-change deadline remain. For authentication, access, or usage errors, check the key, project permissions, and limits before judging the writing. Requests incur charges, so avoid an unbounded retry loop.

Terra supports Responses and Chat Completions, but tool calling requires the Responses API. Its API reasoning levels are none, low, medium, high, xhigh, and max, with medium as the API default. The example explicitly uses low; do not infer that every app defaults to medium. Web search and computer use require additional tools and an execution environment.

Practical requests for everyday work

Match photos to a catalog: Give photographs filenames and supply product codes, colors, and dimensions. Ask for a table of candidate matches, supporting evidence, and uncertain items, leaving unresolved products unconfirmed. Check dimensions against records when a photo cannot establish them, and verify product codes before publishing.

Update an application: Specify both the goal and what to preserve: “Change only this form's title and date, save it, and reopen it to verify the result.” Start with a test record and keep previous values so recovery is possible. Set a clear review boundary before actions such as purchases or sending messages.

Revise a blog post or proposal: Provide sources and the draft, then name the reader, desired length, and claims, numbers, and exceptions to retain. Compare the revision against the original for lost evidence links or qualifications. Shorter does not automatically mean more accurate.

Fix code: Supply reproduction steps, the allowed file scope, expected behavior, and verification commands. Review the diff and checks afterward, including unrelated file changes. When the scope or design uncertainty grows, compare Sol and Astra under the same conditions.

Specifications and limits to check

The official model specifications list a 1,050,000-token context window, a maximum output of 128,000 tokens, and a June 1, 2026 knowledge cutoff. Context is the information a request can accommodate; it is not the same as the output limit. Product-level allowances may differ.

Terra accepts and produces text, accepts images, and does not directly support audio or video modalities. Connecting an image-generation tool is also different from native image output. For recent news or changed policies, supply current primary material or use an appropriately connected search tool.

Before expanding adoption, choose one representative task and define completion. For writing, check required facts and exceptions; for screen work, re-query saved values; for code, confirm the fix and regression checks. These criteria make it easier to judge whether Terra is suitable for your actual work.