On October 26, 2026, OpenAI introduced GPT-6.1 Terra, a model built for office work and for using a computer on the user's behalf: writing, reading documents, filling in forms. It is available in ChatGPT on the Plus, Pro, Enterprise and Edu plans, and in the API at $1 per million input tokens.
What we know
- What it is, per OpenAI: a "fast and efficient" model that combines visual reasoning, computer use and more natural writing. OpenAI compares it mainly with GPT-5.6 Terra, the previous model in that family.
- Who gets it: from launch day, Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. In the API it is called gpt-6.1-terra. The announcement doesn't mention the free plan.
- API pricing: $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. OpenAI says that is 50% less than GPT-5.6 Terra and GPT-6.1 Sol.
- Performance, on OpenAI's tests: on Agents' Last Exam at medium reasoning effort it scores 15 points higher than GPT-5.6 Terra at about a third of the cost. On Terminal-Bench 4.0 at maximum effort it solves 50.3% of tasks, versus 21.5% for its predecessor.
- Factual errors: on an internal OpenAI test with misleading prompts, the error rate at low effort fell from 53.6% to 16.7%; at maximum effort it ended at 4.3%.
- Limits OpenAI acknowledges: for the hardest coding work it still recommends GPT-6.1 Sol or GPT-6 Astra. On safety, it says the model reaches the critical cybersecurity threshold in its Preparedness Framework and keeps the safeguards built for GPT-6 Astra.
What changes and what doesn't
The cost of automating high-volume tasks changes. At the published prices, processing 1,000 invoices a month, at roughly 3,000 input tokens and 500 output tokens each, would cost about $5.50 ($3 input, $2.50 output). That figure is this article's estimate: real usage depends on the document, the reasoning effort and how much cached input you reuse.
What doesn't change is that every performance figure comes from OpenAI, on its own tests, some internal, with competitor comparisons the company chose. A 4.3% error rate in the best setting is still one wrong answer in 25: anything that reaches a customer or an invoice needs a review.
How to tell if it affects you
- If you pay for Plus or a business plan, you can pick it in ChatGPT now. Try it on a real task, such as summarizing a contract or drafting a proposal, and compare it with the model you use today.
- If you use the API, gather 20 real cases (emails, product listings, invoices), run them through your current model and gpt-6.1-terra, and compare quality, time and cost before changing anything in production.
- Do your own math. Multiply your monthly documents by the input and output tokens for each and apply the prices above. If most of it is repeated input, caching changes the total a lot.
- Review your web forms. Models like this fill in forms for users. Clear labels, readable field names and text error messages help an agent complete a booking or an order without mistakes.
- Keep sensitive steps human. Payments, signatures and personal data still need a person's review, however much the model improves on benchmarks.
Related: OpenAI's GPT-6 Model Guide: Which Model to Pick, and Cost
Sources
Updates: if OpenAI opens it to the free plan or changes its pricing, it will be added here with a link.