← Back to the blog AI Consulting · Toledo, Ohio

OpenAI's New Terra Model Fills Out Forms and Processes Invoices. Test It on One Chore First.

Jayson Hines · October 27, 2026 · 4 min read

OpenAI launched a model called GPT-6.1 Terra on Monday. Sam Altman posted about it on X and said they're especially excited about "this new era of computer use." Translation: it's built to click around in software the way a person does. OpenAI's own examples are filling out forms, processing invoices, editing video and reading technical drawings.

It's also cheaper. The API price is $1 per million input tokens and $5 per million output tokens, which is half what OpenAI lists for GPT-6.1 Sol. Paying ChatGPT customers on Plus, Pro, Enterprise and Edu can use it through ChatGPT Work and Codex.

What OpenAI says, and what nobody has checked

OpenAI claims Terra finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. It also says the error rate on hard prompts at low reasoning effort dropped to 16.7%, down from 53.6% on the older GPT-5.6 Terra. Those numbers came from OpenAI. I haven't seen anyone outside the company confirm them.

Read that error rate again, too. Even on OpenAI's good number, roughly one hard task in six goes wrong. OpenAI itself says Terra isn't meant to replace its top models everywhere, and that for the most complex coding work Sol and Astra are still the better pick. The company also recommends testing on your own workloads before you move anything over.

Why this one matters to a small shop

Most of what I see AI do in Toledo businesses is writing. Emails, quotes, social posts. Computer use is a different animal because the AI is touching the same screens you touch. Your invoicing program, your bank portal, your scheduling software, the form on the state website.

A bad paragraph is easy to catch. A wrong number typed into an invoice, or a form submitted with the wrong date, can sit there for a month before anybody notices. And since Terra is cheap, there's going to be a lot of pressure to point it at the boring paperwork first, because that's the work everybody hates.

Pick one chore and run it twice

Don't start with your money. Pick one repetitive job where a mistake costs you nothing, like entering the same kind of info into a spreadsheet or filling in a vendor form you submit every week. Have the AI do it while you watch, then check every field yourself.

Do it again the next day with a different batch. If it's clean twice, you've learned something real about your own work instead of a number from a launch post. If it flubs something, you found out on a chore that didn't matter.

Whatever you try, keep it away from payments and anything that sends something to a customer until you've watched it get things right for a few weeks. And decide ahead of time who looks over the finished work. Somebody has to own that check, or it won't happen. If your team already has a few ideas for where AI could take over screen work, I covered what a small business AI policy needs to say, and that's a good place to write the rule down.

A cheaper model that can run your software is useful. Just make it earn the job before you hand it the keys.

I run AI workshops and one-on-one AI consultations for companies around Toledo, Northwest Ohio, and Southeast Michigan, and picking the first safe chore to hand an AI is a common starting point. If you want help finding yours and setting up a check that actually gets done, email me and we'll set up a time.

Email Jayson