OpenAI launched GPT‑6.1 Terra on 26 October 2026, an artificial‑intelligence system that can watch a computer screen, click buttons, type into fields and move between applications to carry out a user‑specified job, and the service is now being rolled out to paying ChatGPT customers.
What is “computer use”?
Unlike typical chatbots that only answer queries, a computer‑use model performs actions. Users describe a task—such as copying invoice data into accounting software or completing a lengthy form—and the AI manipulates the graphical interface just as a human would, reading on‑screen information, selecting options and entering text.
Who can access the model?
According to OpenAI, GPT‑6.1 Terra is being offered to subscribers of ChatGPT Plus, Pro, Enterprise and Edu plans within the ChatGPT Work suite, as well as to developers using the Codex coding tool. Free accounts are excluded. The cheapest paid tier, ChatGPT Plus, costs $20 per month. Developers may also call the model via OpenAI’s paid API at half the price of the predecessor model it replaces.
What can the AI accomplish?
OpenAI showcases several use‑cases: extracting data from invoices and entering it into bookkeeping software, auto‑filling online forms, drafting business proposals, interpreting technical drawings and even performing basic video edits. The company claims these operations are completed more than three times faster than with its larger, earlier models.
How reliable is the system?
In internal testing with prompts designed to provoke mistakes, Terra produced incorrect results 16.7 % of the time on its fastest, lowest‑reasoning setting—down from 53.6 % for the model it supersedes. On the most careful setting, the error rate fell to 4.3 %, meaning roughly one error in twenty‑four attempts. These figures are based solely on OpenAI’s own evaluations and have not been independently verified.
What should users consider before granting access?
- Visibility: The model can see everything displayed on the screen, including bank balances, messages and personal data.
- Mistakes become actions: An erroneous answer in a chat is easy to dismiss, but a wrong click can trigger a payment, deletion or data submission.
- Enable safeguards: Look for settings that require confirmation before the AI sends, purchases or deletes anything, and keep those safeguards active.
- Start with low‑risk tasks: Begin by letting Terra sort files or organize folders before entrusting it with financial transactions.
Is the upgrade worth the cost?
For small businesses burdened with repetitive data‑entry work, the time saved could be significant, provided a human reviews the output. For most individual users, the free version of ChatGPT already handles everyday queries, and the computer‑use capability alone may not justify the subscription fee.
How to begin using Terra?
- If you already pay for ChatGPT, select a simple, low‑stakes task and monitor each step the AI takes.
- Activate any “ask‑before‑acting” options before assigning real‑world work.
- Never allow the model to manage passwords, make payments or perform irreversible actions without supervision.
Primary source: OpenAI’s announcement (https://openai.com/index/introducing-gpt-6-1-terra/).