GPT-6.1 Terra: what OpenAI's new model means for SMEs
On October 26, 2026, OpenAI released GPT-6.1 Terra: a fast, comparatively affordable model that understands screens and can operate software itself. If you have seen it called “ChatGPT Terra 6.1”, this is the same model. For small and medium-sized businesses, one question matters most: can GPT-6.1 Terra take over routine screen work – filling in forms, moving data between programs, checking entries?
This article puts the official information into context: features, use cases, pricing and limits. Note up front: Lyron has not yet tested GPT-6.1 Terra; all performance figures come from OpenAI.
In short
- New mid-range model: GPT-6.1 Terra succeeds GPT-5.6 Terra and sits between GPT-6 Luna and the stronger GPT-6.1 Sol and GPT-6 Astra.
- Focus on computer use: the model recognises screen content and can click, type and fill in forms.
- Available: in ChatGPT Work and Codex on paid plans and via the API, not in regular Chat.
- Cheaper: $1 per million input and $5 per million output tokens – per OpenAI, 50% less than GPT-5.6 Terra and GPT-6.1 Sol.
- Control remains a must: test accounts, approval before submitting and a data protection review come before any production use.
What is GPT-6.1 Terra?
OpenAI positions GPT-6.1 Terra as a fast, efficient model that pairs detailed image understanding with precise computer control and clearer writing – a clear step up from GPT-5.6 Terra. Typical tasks, per OpenAI: drafting proposals, interpreting technical drawings, filling out forms, editing videos.
Terra sits in the middle of OpenAI's lineup. Per the developer docs, all four handle up to 1.05 million tokens of context and 128,000 output tokens; they differ mainly in capability and price:
| Model | OpenAI's positioning | USD per 1M tokens (input / output) |
|---|---|---|
| GPT-6 Luna | most economical model for focused, high-volume tasks | 0.10 / 0.50 |
| GPT-6.1 Terra | fast and efficient: visual reasoning, computer use, clear writing | 1 / 5 |
| GPT-6.1 Sol | near-Astra results on complex work at lower cost | 2 / 10 |
| GPT-6 Astra | most capable model for the most demanding work | 10 / 50 |
Computer use: where GPT-6.1 Terra helps SMEs day to day
Computer use means the model looks at screenshots, decides the next step and clicks, scrolls or types – like a person at the screen. In ChatGPT, this runs in supported regions through the desktop app's Computer Use plugin; ChatGPT asks for permission before it operates a program. Developers use the API.
This is useful wherever data is retyped by hand because no interface exists:
- Forms in portals: transferring details from emails or PDFs into supplier, government or customer portals.
- Data between programs: entering values into older industry software without an API, or reconciling two systems.
- Capturing documents: OpenAI itself cites reading incoming invoices and entering their details into accounting software.
Fictional example: Every week, an electrical contractor with twelve employees copies details from PDF orders into a grid operator's web portal that has no interface. An assistant based on GPT-6.1 Terra could read the PDF, fill in the form and stop before submitting; an employee checks the entries and clicks “Submit” herself. The scenario is made up – no customer project, no measured results.
OpenAI itself advises using a program's own integration for recurring tasks where one exists. Where an API is available, a workflow for business process automation, for example with n8n, is usually more stable than operating the screen. Our page on AI agents explains how assistants work within clear rules.
What OpenAI promises: benchmarks and speed
All figures in this section are OpenAI's own; independent tests were not available at publication.
- Computer tasks: on Agents' Last Exam, a benchmark of professional computer tasks, Terra scores 15 percentage points above GPT-5.6 Terra at medium reasoning effort, at about a third of the cost. OpenAI says it also performs on par with GPT-6 Astra at a tenth of Astra's price per token, while finishing tasks over three times faster than Astra and GPT-6.1 Sol.
- Desktop apps: on OSWorld 2.1, Terra at medium reasoning effort beats GPT-5.6 Terra's best score – at a fraction of the cost.
- Factuality: in OpenAI's internal test of error-inducing prompts, the error rate fell from 53.6% to 16.7% at low reasoning effort, and to 4.3% at maximum effort.
- Coding: on Terminal-Bench 4.0, Terra solves 50.3% of tasks at maximum reasoning effort, versus 21.5% for GPT-5.6 Terra. For the most complex coding tasks, OpenAI still recommends Sol or Astra.
Good scores do not mean error-free work – plan to check results from day one.
Availability and pricing of GPT-6.1 Terra
In ChatGPT, GPT-6.1 Terra has been available in ChatGPT Work and Codex since October 26, 2026 – not in regular Chat. OpenAI's sources differ on plans: the announcement lists Plus, Pro, Enterprise and Edu, the changelog adds Business; neither mentions the free version. In Enterprise workspaces, administrators control access. Per OpenAI, a plan's usage allowance covers more tasks with Terra than with GPT-5.6 Terra, and substantially more than with Sol or Astra.
In the API, the model is called gpt-6.1-terra. Standard prices per million tokens: $1 for input, $0.05 for cached input and $5 for output – per OpenAI, 50% below GPT-5.6 Terra and GPT-6.1 Sol. Batch and Flex cost half; requests over 272,000 input tokens cost more. Computer use requires the Responses API; screenshots count as image input.
Limits and risks: permissions, approvals, data protection, costs
A model that operates software acts with the rights of the signed-in account. Settle five points before any use:
- Permissions: if ChatGPT uses your browser, it can act on sites where you are signed in – and those sites may treat the clicks as yours. Test accounts with minimal rights and a separate browser limit the risk; administrators can decide which programs computer use may access.
- Approvals: OpenAI recommends having a person confirm purchases, data transmissions and hard-to-reverse changes – even typing sensitive data into a form counts as a transmission. See our guide to permissions and approvals for AI agents.
- Manipulated content: web pages and documents can hide instructions (prompt injection). OpenAI reports strong resistance for Terra in its own tests, but advises treating screen content as untrusted.
- Data protection and GDPR: screenshots show everything visible, including customer data. In ChatGPT, your account's data controls apply; via the API, OpenAI does not use data for training by default. API customers can also get EU data residency for Terra, but only with OpenAI's approval and at a 10% premium. Settle data processing agreements, data minimisation and deletion periods before using real personal data.
- Costs: low token prices say little about the cost per task, because every step produces screenshots. OpenAI recommends limits on steps, time and budget.
This article is practical guidance and does not replace legal advice.
How to test GPT-6.1 Terra safely: five steps
- Choose a small task: for example, moving data from a PDF into a web form – no payments, deletions or outbound messages.
- Set up a test environment: a test account with minimal rights, a separate browser or virtual machine, test data instead of customer data.
- Define approval points: the model stops before “Save” or “Send”, and a person confirms. Use “Always allow” only for non-critical programs.
- Check and measure: compare every entry with its source; log errors, time and cost per task.
- Decide deliberately: if there is an interface, a workflow is usually more robust. Computer use fits where none exists – once data protection is settled.
Frequently asked questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is an OpenAI model released on October 26, 2026 that understands screens and can operate software through computer use. It succeeds GPT-5.6 Terra and sits between GPT-6 Luna and GPT-6.1 Sol.
Is GPT-6.1 Terra available in the free version of ChatGPT?
No. According to OpenAI, the model is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu; the changelog also lists Business. It cannot be selected in regular Chat.
How much does GPT-6.1 Terra cost in the API?
Standard pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Batch and Flex processing cost half as much.
Should GPT-6.1 Terra be allowed to submit forms on its own?
Technically, the model can fill in and submit forms. But OpenAI recommends having a person confirm data transmissions, purchases and hard-to-reverse changes: the model prepares, a person approves.
Sources
As of October 26, 2026. Figures and model details come from OpenAI; the assessment, example and test steps are Lyron's own.
- OpenAI: Introducing GPT-6.1 Terra – announcement with benchmarks, pricing and availability.
- OpenAI API docs: GPT-6.1 Terra and Models – pricing, context window and lineup.
- ChatGPT & Codex changelog – entry of October 26, 2026 on plans.
- ChatGPT Learn: Computer Use and OpenAI API docs: Computer use – permissions, approvals and safe operation.
- OpenAI API docs: Data controls – data usage and data residency.
- 9to5Mac and BleepingComputer – reports of October 26, 2026.
Want to know whether computer use makes sense for your business?
In a free initial consultation, we review your recurring screen work: where an interface or workflow is enough, where an AI agent with approvals fits and what a safe test looks like.
Free Consultation