OpenAI released GPT-6.1 Terra on Monday, Oct. 26. It is an AI model built for “computer use”: it looks at a screen, then clicks, types and moves between apps to finish a task for you. It is rolling out to paying ChatGPT subscribers now. On OpenAI’s own test of prompts designed to trip it up, it still gets about one in six wrong at its fastest setting.
What is “computer use”?
Most chatbots answer questions. A computer-use model does things. You describe a task, such as copying invoice details into accounting software or filling out a long form, and the AI works the apps the way a person would: it reads what is on screen, clicks buttons and types into fields.
Who can use it?
OpenAI says GPT-6.1 Terra is rolling out to ChatGPT Plus, Pro, Enterprise and Edu users in ChatGPT Work and in Codex, OpenAI’s coding tool. Free accounts do not get it. ChatGPT Plus, the cheapest plan included, costs $20 a month. Developers can also use it through OpenAI’s paid API, where OpenAI says it costs half as much as the model it replaces.
What can it do?
OpenAI’s examples include reading invoices and entering them into accounting software, filling out forms, drafting proposals, reading technical drawings and editing video. OpenAI says it finishes these tasks more than three times faster than its larger models.
How often does it get things wrong?
On OpenAI’s internal test of prompts designed to cause mistakes, Terra gave a wrong answer 16.7% of the time at its lowest reasoning setting, the fastest one,, down from 53.6% for the model it replaces. At its most careful setting the error rate fell to 4.3%. That is a big improvement. It still means about one wrong answer in six at the fast setting, and when an AI is clicking “submit” for you, a wrong answer becomes a wrong action. These numbers come from OpenAI’s own testing and have not been checked independently.
What should I think about before letting AI use my accounts?
- It sees what you see. Bank balances, messages and personal details on screen are visible to the model while it works.
- Mistakes become actions. A wrong answer in a chat is easy to ignore. A wrong click can send, delete or pay.
- Look for a stop button. Check for settings that make it ask before it sends, buys or deletes anything, and keep them on.
- Start small. Sorting files is a safer first test than paying bills.
Is it worth paying for?
For someone running a small business with lots of repetitive data entry, it could save real hours, as long as a person checks the result. For most people, the free version of ChatGPT still handles everyday questions, and computer use alone is not a reason to upgrade.
What to do next
- If you already pay for ChatGPT, try Terra on one boring, low-risk task and watch every step.
- Turn on any ask-before-acting settings before giving it real work.
- Never let it handle passwords, payments or anything you cannot undo.
Primary source: OpenAI’s announcement
Sources: OpenAI, “Introducing GPT-6.1 Terra,” Oct. 26, 2026; OpenAI on X, Oct. 26, 2026; BleepingComputer, Oct. 26, 2026; The Next Web, Oct. 26, 2026.
