GPT-6.1 Terra is a new OpenAI model that uses your computer for you, so a small business can hand it screen work like forms and data entry. It’s live now for ChatGPT Plus, Pro, Enterprise and Edu users, and you should start it on one low-risk task while you watch.
OpenAI announced GPT-6.1 Terra on October 26 (BleepingComputer’s write-up has the same facts). Here’s what it is, what I’d give it first, and when you’re better off with a proper automation.
What is GPT-6.1 Terra?
It’s an AI model that looks at your screen and acts on it. It reads what’s there, then clicks, types and moves between apps, the way you would.
OpenAI calls this “computer use.” Its own examples include filling out forms, drafting proposals, reading technical drawings and editing videos.
It’s the cheaper, faster model in OpenAI’s lineup, not the top one. OpenAI says it finishes tasks over three times faster than its bigger models, GPT-6 Astra and GPT-6.1 Sol.
Who can use it, and what does it cost?
You can use it today if you pay for ChatGPT Plus, Pro, Enterprise or Edu. It runs inside ChatGPT Work and Codex. The announcement doesn’t list the free plan.
OpenAI doesn’t give a separate price for owners. It says your existing plan allowance covers more tasks with Terra than with its other models.
If you have a developer, they can call it through the API as gpt-6.1-terra. That costs $1 per million input tokens and $5 per million output tokens, half what the older GPT-5.6 Terra cost.
What could it do in your business?
Think about the jobs where someone sits at a screen and copies, clicks and pastes. Those are the ones to look at.
- Old software with no connection to anything. You have a tool that doesn’t talk to your other systems. Terra can click through it like a person would.
- Copying data between systems. OpenAI’s own example is reading incoming invoices and entering them into accounting software. The same idea works for orders, leads or job details.
- Filling the same web forms. Supplier portals, permit sites, warranty claims. Anything where you type the same kinds of details into boxes again and again.
- Pulling reports from portals. Logging into a dashboard, downloading last month’s numbers and putting them in a sheet.
How do I try it safely?
Treat it like a new hire on day one. You wouldn’t hand a new hire your bank login.
- Pick a low-risk internal task. Something where a mistake costs you a few minutes, not a customer. Pulling a report is a good first one.
- Watch the first few runs. Sit with it while it works. You’ll see fast where it gets confused.
- Keep it out of sensitive logins at first. No banking, no payroll, no customer data until you trust it on easier work.
- Check its work every time early on. Compare its output to what you’d have done. Loosen up only once it’s been right for a while.
I wrote more on this in before you hand AI an important job.
If you’d rather have someone set this up safely, Book a free call.
What it can’t do yet
It still makes mistakes. On OpenAI’s own test of tricky prompts, its error rate was 16.7% at the lowest effort setting and 4.3% at the highest. Those are OpenAI’s numbers, not an outside test.
OpenAI also says its bigger models are still the better pick for the most complex coding work. And it rates Terra at its “Critical” level for cybersecurity ability, meaning it treats the model as powerful enough to need extra safeguards.
OpenAI says Terra showed “strong resistance” to prompt injection, which is hidden text on a page that tries to steer the AI. Strong isn’t perfect, so keep watching what it does on sites you don’t control.
When should you build a proper automation instead?
Computer use is great for jobs that are messy, rare or stuck in software with no other way in. It’s slower and less predictable than a direct connection between two systems.
If a job runs every day the same way, like syncing new orders into your books, a proper integration is usually cheaper and more reliable. It doesn’t need to look at a screen, so it doesn’t get confused when a button moves.
My rule is to use Terra to figure out a task and handle the odd ones. Once the task runs daily, build it properly.
Anthropic has its own agent for small businesses too. Here’s my take on Claude Duet if you want to compare.
FAQ
What is GPT-6.1 Terra?
It’s an OpenAI model released October 26, 2026 that can see your screen and use apps for you. It clicks, types and fills in forms like a person would.
Is GPT-6.1 Terra free?
OpenAI lists it for Plus, Pro, Enterprise and Edu users, not the free plan. Developers pay $1 per million input tokens and $5 per million output tokens through the API.
Is it safe to give GPT-6.1 Terra my logins?
Not at first. Start with a low-risk task, watch it work, and keep it away from banking, payroll and customer data until it’s proven itself.
Should I use Terra or build an automation?
Use Terra for messy or occasional screen work. For jobs that run the same way every day, a direct integration is usually cheaper and more reliable.