TL;DR
What is GPT-6.1 Terra?
GPT-6.1 Terra is OpenAI's new mid-tier AI model, built for "computer use": it reads what's on your screen and takes actions across apps for you. OpenAI released it on Monday, October 26, 2026. It's rolling out to ChatGPT Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers in the API as gpt-6.1-terra, priced at US$1 per 1M input tokens and US$5 per 1M output tokens. That's half the standard API price of GPT-5.6 Terra, and it's also on Amazon Bedrock and Microsoft Foundry. All speed and benchmark numbers so far come from OpenAI itself.
What happened
OpenAI announced GPT-6.1 Terra on Monday, October 26, 2026, posting on X at 2:06 PM ET. It pitches the model as "fast and capable computer use at a lower price." CEO Sam Altman followed a couple of hours later, calling it the start of "this new era of computer use."
Terra is OpenAI's mid-tier option. OpenAI compares it with GPT-6 Astra, which costs ten times as much per token, and with GPT-6.1 Sol. OpenAI's own example of the kind of work it's for is reading invoices into accounting software. It's also live on Amazon's and Microsoft's cloud AI platforms.
What GPT-6.1 Terra does
These are OpenAI's descriptions and claims:
- Reads your screen. OpenAI says it "understands what's on screen," so it can see the app or web page you're using.
- Acts across apps. It can click, type and move between programs to finish a task, such as copying invoice details into accounting software.
- Writes naturally. OpenAI pitches it as a model for your whole workday, including writing.
- Holds a lot at once. It has a 1M-token context window, enough to keep long documents and long tasks in view.
- Finishes faster. OpenAI says it completes tasks over 3x faster on average than GPT-6 Astra and GPT-6.1 Sol.
- Makes fewer mistakes. OpenAI says its error rate at low effort fell from 53.6% to 16.7%, and to 4.3% at max effort.
How much does GPT-6.1 Terra cost?
If you pay for ChatGPT, it comes with your existing Plus, Pro, Enterprise or Edu plan. OpenAI hasn't announced a separate charge. If you build apps on OpenAI's API, you pay per token:
| What you pay for | GPT-6.1 Terra API price |
|---|---|
| Input | US$1 per 1M tokens |
| Cached input | US$0.05 per 1M tokens |
| Output | US$5 per 1M tokens |
Who can use it, and what it costs
| Who | Where | What it costs |
|---|---|---|
| ChatGPT Plus and Pro users | ChatGPT Work and Codex | Included in your plan, rolling out now |
| ChatGPT Enterprise and Edu users | ChatGPT Work and Codex | Included in your plan, rolling out now |
| Developers | OpenAI API, model name gpt-6.1-terra | US$1 input / US$0.05 cached input / US$5 output per 1M tokens |
| Amazon cloud customers | Amazon Bedrock | Billed through AWS (see Bedrock pricing) |
| Microsoft cloud customers | Microsoft Foundry | Billed through Microsoft (see Foundry pricing) |
"Rolling out" means not everyone sees it on day one. If it isn't in your model menu yet, check again in a day or two.
Is it safe to let GPT-6.1 Terra use my computer?
OpenAI rates GPT-6.1 Terra at the "Critical" cybersecurity level under its Preparedness Framework, its internal scale for how much harm a model could do in the wrong hands. OpenAI says that rating comes with extra safeguards. For you, it's a signal of how capable the model is at technical work, which cuts both ways.
The bigger everyday risk is simpler. An AI that can click "Pay," "Send" or "Delete" can also click them by mistake, even at a 4.3% error rate. So give it a narrow job, keep it away from banking and email until you trust it, and approve anything that moves money or leaves the building. We set out five simple rules in our Practical Nerdy View on AI computer use.
The jargon, in one line each
| Term | What it means |
|---|---|
| Computer use | An AI that looks at your screen and clicks, types and switches apps to do a task for you |
| Tokens | The small chunks of text AI models read and write |
| Cached input | Text the model has already seen recently (like the same instructions every time), billed at a big discount |
| Context window | How much the model can keep in mind at once, counted in tokens |
| Benchmark | A standard set of test tasks used to score and compare AI models |
| Agents' Last Exam / Terminal-Bench 4.0 | Benchmarks for AI agents doing real work in office software and in a command-line terminal |
| Preparedness Framework "Critical" cyber rating | A high-risk level on OpenAI's safety scale for hacking-related abilities, which comes with extra safeguards |
| API | The connection developers use to plug an AI model into their own apps, paid per token |
| ChatGPT Work / Codex | OpenAI's ChatGPT workspace for everyday work, and its AI coding tool |
How it compares
| GPT-6.1 Terra | GPT-5.6 Terra (previous mid-tier) | GPT-6 Astra (pricier model) | |
|---|---|---|---|
| API price | US$1 input / US$5 output per 1M tokens | Twice Terra 6.1's standard prices | Terra 6.1 costs one-tenth as much per token |
| Agents' Last Exam | About 15 points higher than GPT-5.6 Terra | Baseline, at about 3x Terra 6.1's cost | Matches GPT-6.1 Terra |
| Speed | Over 3x faster on average than Astra | Not given | Slower (see left) |
| Terminal-Bench 4.0 | 50.3% (up from 21.5%) | Not given | Not given |
| Context window | 1M tokens | Not given | Not given |
X reactions
OpenAI announced the model on its official account, and Sam Altman added his own short note that afternoon.
Our take
Price is the real news here. AI that clicks around your computer has been a pricey demo for most small businesses, and halving the cost moves it into "worth a try on a Tuesday" territory. OpenAI's invoice example is well chosen, because typing numbers from a PDF into accounting software is exactly the job nobody misses.
The benchmark claims are big, and they're all OpenAI's own. A model that matches GPT-6 Astra at a tenth of the price would be a bargain, so wait for outside testers before you bet a workflow on it.
Our advice: pick one boring admin chore, like entering a week of invoices, and let Terra do it while you approve every entry before it's saved. Keep it away from payments, email and anything you can't undo. If it gets a week of invoices right, give it a second chore. Our five rules for AI computer use cover the rest.
— Artificially Nerdy
What to do this week
- If you pay for ChatGPT Plus or Pro: look for GPT-6.1 Terra in ChatGPT Work. When it appears, give it one repetitive task, like entering five invoices into your accounting software, and check every entry before you save. For more chores like this, see our AI admin hacks.
- If you run a small team on Enterprise or Edu: agree on a short list of tasks Terra may do and the actions that always need a human click, such as payments, sending email and deleting files. Our seven rules for keeping an AI agent on a short leash help you set those limits.
- If you build apps or automations on OpenAI's API: try swapping in
gpt-6.1-terraon one workflow and compare your bill. Reuse the same instructions each run, since cached input costs US$0.05 per 1M tokens. - If your business runs on AWS or Microsoft Azure: you can try it in Amazon Bedrock or Microsoft Foundry without a new OpenAI account.
- If you're new to AI: you don't need this yet. Read our rules for AI computer use first, so you know what to switch off before you switch it on.
FAQ
What is GPT-6.1 Terra?
GPT-6.1 Terra is a mid-tier OpenAI model released on October 26, 2026, built for computer use. It reads what's on your screen and takes actions across apps, and OpenAI says it's faster and cheaper than earlier models of its kind.
How much does GPT-6.1 Terra cost?
In the API, it costs US$1 per 1M input tokens, US$0.05 per 1M cached input tokens and US$5 per 1M output tokens, before tax (checked Oct 27, 2026). That's half GPT-5.6 Terra's standard price. ChatGPT Plus, Pro, Enterprise and Edu users get it as part of their plan.
Can I use GPT-6.1 Terra on the free ChatGPT plan?
OpenAI lists Plus, Pro, Enterprise and Edu as the ChatGPT plans getting it. Free users aren't on that list.
Is GPT-6.1 Terra better than GPT-6 Astra?
OpenAI says it matches GPT-6 Astra on Agents' Last Exam, a test of agents using work software, at one-tenth the price per token, and finishes tasks over 3x faster on average. These are OpenAI's own figures.
Is it safe to let GPT-6.1 Terra control my computer?
OpenAI rates it "Critical" for cybersecurity under its Preparedness Framework and says it added extra safeguards. For everyday use, start with one low-risk task, keep it away from payments and email, and approve anything you can't undo.
Where else can I get GPT-6.1 Terra?
Besides ChatGPT and the OpenAI API, it's available on Amazon Bedrock and Microsoft Foundry.
Sources
- OpenAI: Introducing GPT-6.1 Terra — the official announcement with availability, prices, benchmark claims and the Preparedness Framework rating.
- OpenAI API docs: GPT-6.1 Terra — the
gpt-6.1-terramodel page with token prices and the 1M-token context window. - AWS: OpenAI GPT-6.1 Terra on Amazon Bedrock — Amazon's availability announcement.
- Microsoft: GPT-6.1 Terra now generally available in Microsoft Foundry — Microsoft's availability announcement.
- OpenAI on X — the October 26, 2026, launch post (2:06 PM ET).
- Sam Altman on X — Altman's October 26, 2026, post (4:37 PM ET).
- The Verge: GPT-6.1 Terra is very good at computer use, OpenAI says — news coverage of the launch.
