# GPT-6.1 Terra in Codex: What Changed and Who Should Care > OpenAI added GPT-6.1 Terra to Codex on 26 Oct 2026: a cheaper, faster model built for screens and images. What it is, what it costs, and how to test it. Canonical: https://bridge-terminal.com/atlas/blog/gpt-6-1-terra-codex-what-changed/ Last updated: 2026-10-27 Published 27 October 2026 · Plain English · What's new · 6 min read ## The short version - On 26 October 2026 OpenAI made GPT-6.1 Terra, a fast, efficient model, available in Codex and ChatGPT Work on Plus, Pro, Business, Enterprise and Edu plans. - OpenAI describes Terra as built for visual reasoning, precise computer use and clear writing, so it suits work that involves screens and images. - In the API Terra costs $1 per million input tokens and $5 per million output tokens. OpenAI's pages do not claim it is the best model for the hardest coding. - A new model is a reason to run a small, measured trial on your own tasks, not to switch the whole team on day one. - Atlas runs on Claude Code, so it does not use Terra; that is a fact about Atlas's design, not a verdict on the model. GPT-6.1 Terra is a new OpenAI model that became available in Codex, OpenAI's AI coding agent, on 26 October 2026. OpenAI describes it as a fast, efficient model for professional work, strong at understanding images and using a computer on your behalf. If your team uses Codex, whether Terra is worth trying depends on the kind of work you give the agent. ## What is GPT-6.1 Terra? A "model" is the AI brain behind a coding agent. The agent (Codex) is the tool you talk to. The model is what reads your request, thinks, and decides what to do. Different models trade off speed, price and ability. OpenAI's changelog entry for 26 October 2026 says GPT-6.1 Terra is available in ChatGPT Work and Codex on the Plus, Pro, Business, Enterprise and Edu plans. It is not available in Chat, and it does not support UAE inference residency (a setting that keeps processing inside the United Arab Emirates). For Enterprise customers, access depends on workspace availability and administrator controls. OpenAI's own description is "a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing." In plain terms, it can look at a picture or a screen, work out what is on it, and then act across your browser, desktop and apps. OpenAI's examples are not about code at all: matching supplier product photos to catalog entries, and checking campaign assets for cropping and text placement. It also says Terra writes in a clear, direct style, with the result first and the supporting details explained in plain language. ## What actually changed on 26 October? Three things, all from OpenAI's own pages: - **A new choice in the model list.** In the ChatGPT desktop app, you find Terra under "Advanced" in the model list. In ChatGPT Work on the web it is in the same list, and Work on mobile supports it too. - **A published price.** OpenAI's API page lists the model as gpt-6.1-terra at $1 per million input tokens and $5 per million output tokens, with cached input (text the model has already been sent and can reuse) at $0.05 per million. A "token" is a small chunk of text; you pay for what the model reads and writes. The context window, meaning how much it can keep in view at once, is 1,050,000 tokens. - **Setup rules for computer use.** OpenAI notes that browser use and computer use each have their own setup and access requirements. Picking the model does not switch those on by itself. News coverage, such as 9to5Mac on 26 October, reports that OpenAI says Terra costs 50% less than GPT-5.6 Terra, the model it replaces, and that OpenAI has four model classes, Luna, Terra, Sol and Astra, ordered from lowest to highest price and capability, with Terra in the middle. Those comparisons come from OpenAI, reported by a third party, so treat them as the maker's claim rather than a test on your own work. ## Is Terra a coding model? Not mainly, going by how OpenAI describes it. The changelog and the model page talk about images, screens, computer use and writing. Neither page gives coding benchmarks, and neither says Terra is the best choice for the hardest programming tasks. The API model page does not mention Codex at all; the Codex availability comes from the changelog entry and from 9to5Mac. That does not mean Terra is poor at code. It means OpenAI has not said, on these pages, that it is good at it, and a careful team should not assume either way. Compare this with the more usual "new model" story, where the headline is better code. Here the headline is a lower price and better handling of what is on a screen. ## What might it mean for a team building software? Think about the tasks where an agent has to look at something rather than only read code. These are ideas to test, not claims OpenAI makes about Terra: - **Checking a screen against a design.** Does the page the agent built look like the mock-up? A model that is good with images may judge this better. - **Walking through an app like a user.** Computer use means the agent clicks and types in a real interface, which could help with flows that have no automated test yet. - **Reading screenshots in bug reports.** Testers and analysts often attach pictures. A model that understands them might turn them into clearer tickets. - **Writing for people.** Release notes, summaries and explanations of what changed. OpenAI stresses clear writing and results first. For deep changes inside a large codebase, nothing on these pages tells you Terra is the right pick. If cost is your worry, a cheaper model can help on routine work, but a cheaper model that needs two or three attempts can cost more than a dearer one that gets it right the first time. That is why the trial below counts attempts, not only price. ## How should we try it without risking the project? Run a small, honest trial. This works for any new model from any maker. 1. **Pick three to five real tasks** you have already finished: a bug fix, a small feature, a screen to check. Real ones, not demos. 2. **Run each on your current model and on Terra**, with the same written brief, so the only thing that changes is the model. 3. **Measure the same things each time:** did the tests pass, how many tries did it take, how long did it run, how much of your plan or budget did it use, and did a person need to fix anything afterwards. 4. **Decide per kind of work.** You may end up with one model for screen checks and writing, another for heavy code. That is a normal outcome. 5. **Check your admin settings first** if you are on a Business or Enterprise plan, since access depends on your workspace and administrators. Models also come and go. If your team pins a model by name in a script or setting, keep a list of where, so a model change does not quietly break a workflow. ## What about security with computer use? Letting an agent control a browser or desktop is a bigger step than letting it edit files. It can click "send", "delete" or "buy" just as a person can. OpenAI says computer use has its own setup and access requirements, which is a reason to read them before turning it on. Start with test accounts and non-sensitive data, and keep a person approving anything that sends, spends or deletes. ## How does Atlas fit? Atlas is a software team of AI engineers built on Claude Code, and it signs in with your own Claude plan. That means it does not use OpenAI models such as Terra, and this release changes nothing about how Atlas works. What Atlas does do is give each engineer a written role, with the model and effort suited to that role, which is the same idea of matching the model to the job. If you are curious how the pieces fit, see the [Atlas home page](https://bridge-terminal.com/atlas/). Atlas is an independent product. ## The short conclusion GPT-6.1 Terra is a real, dated change for Codex users: available since 26 October 2026, priced at $1 and $5 per million tokens in the API, and aimed at images, screens and clear writing rather than being sold as the strongest coder. Treat it as a candidate for specific jobs, test it on your own finished work, and let the numbers, not the launch post, decide where it goes. ## Questions people ask ### What is GPT-6.1 Terra? It is an OpenAI model that OpenAI calls fast and efficient, built for visual reasoning, precise computer use and clear writing. It became available in Codex and ChatGPT Work on 26 October 2026. ### How much does GPT-6.1 Terra cost? In the API, OpenAI lists $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million. 9to5Mac reports OpenAI says that is 50% less than GPT-5.6 Terra, the model it replaces. ### Who can use GPT-6.1 Terra in Codex? OpenAI says it is available in Codex and ChatGPT Work on Plus, Pro, Business, Enterprise and Edu plans. It is not in Chat, and Enterprise access depends on workspace availability and admin controls. ### Is GPT-6.1 Terra the best model for coding? OpenAI's pages do not say so. They describe it for images, screens, computer use and writing and give no coding benchmarks, so test it on your own tasks before switching. ### Does Atlas use GPT-6.1 Terra? No. Atlas is built on Claude Code and runs on your own Claude plan, so OpenAI models are not part of it. ## About Atlas: Matching the model to the job, on Claude Code Terra is an OpenAI model, so it is not part of Atlas, which runs on Claude Code with your own Claude plan. The idea behind it still applies: Atlas gives each AI engineer a written role with the model and effort suited to that role. Atlas is an independent product. — https://bridge-terminal.com/atlas/ ## Sources - [OpenAI: What's new, October 26–30, 2026 (GPT-6.1 Terra)](https://learn.chatgpt.com/docs/whats-new/october-26-30-2026) - [OpenAI Codex and ChatGPT changelog](https://learn.chatgpt.com/docs/changelog) - [OpenAI API docs: GPT-6.1 Terra model page](https://developers.openai.com/api/docs/models/gpt-6.1-terra) - [9to5Mac: OpenAI upgrades ChatGPT and Codex with new GPT-6.1 Terra model](https://9to5mac.com/2026/10/26/openai-upgrades-chatgpt-and-codex-with-new-gpt-6-1-terra-model/) _Researched and drafted with Claude, then checked against the sources above. Atlas is an independent product, not affiliated with Anthropic or OpenAI._ ## Related - [Codex GPT-5.5 Retirement: What Your Team Should Check](https://bridge-terminal.com/atlas/blog/codex-gpt-5-5-retirement-what-to-check/) - [How to Brief an AI Coding Agent: A Plain-English Guide](https://bridge-terminal.com/atlas/blog/how-to-brief-ai-coding-agent-plain-english/) - [What Is a Flaky Test? A Plain Guide for AI-Built Software](https://bridge-terminal.com/atlas/blog/what-is-a-flaky-test-plain-guide/)