OpenAI has a new model out, and it is not the fancy kind. GPT-6.1 Terra arrived on October 26, 2026, and its whole job is the dull stuff — typing details into form after form, entering invoices, fixing drafts, cutting videos. It works quickly, and it runs at half the price of the model before it. That second part is the real headline.
What Happened
The launch date was Monday, October 26, 2026.
Inside the GPT-6 family, Terra sits in the middle. Luna is the small one, Sol and Astra the big ones. OpenAI calls it mid-tier: more muscle than Luna, a lower bill than the other two.
One small oddity, though: when the first GPT-6 models came out last month, Terra was not among them. The family shipped without it. Only now does it catch up.
Two strengths get pointed at most. Terra reads screens with sharp eyes, and it works the computer with real care — right buttons pressed, text typed where it belongs, app to app without fuss. Its writing is clearer and more natural than the old Terra’s, too.
You can use it today. It is already running inside ChatGPT Work and Codex, and developers can reach it through the API, where it answers to the name gpt-6.1-terra. Access covers Plus, Pro, Enterprise, and Edu users.
If you followed the first GPT-6 launch, this is the missing piece of that family. Read our earlier post on ChatGPT GPT-6 intelligent UI launch for the background.
What GPT-6.1 Terra Can Do
Strip away the marketing and the idea is plain: you give Terra a computer job, and it does the job.
It looks at your screen, works out what it is seeing, and gets moving. A click here. Some typing there. One app closed, another opened, step after step, until the job is done — a whole chain of actions, on its own.
And what sort of work? OpenAI’s examples sound almost dull, honestly, but dull in the best possible way:
- Putting fresh invoices into accounting software.
- Working through long forms that nobody wants to fill.
- Turning rough notes into a proper proposal draft.
- Reading a technical drawing and saying what it shows.
- Giving a hand with video editing.
Ring any bells? If you study or freelance, you have done all of these — and they swallow hours, week after week.
Vision is part of the package too. Hand Terra a chart and it digs out the numbers that matter; hand it a document and it finds the part that counts. Then it can act on what it found. That lifts it above a chatbot that only talks.
The Price: Half the Cost of the Old Model
Now for the money — the part that makes this launch worth your time.
On the API, GPT-6.1 Terra charges $1 per million input tokens. Output costs $5 per million tokens. Cached input? $0.05 per million. Pennies.
Compare the model it replaces: GPT-5.6 Terra ran at roughly twice the price. The bigger sibling, GPT-6.1 Sol, charges $2 in and $10 out — against Sol too, the new Terra is half price.
A side-by-side look:
- GPT-6.1 Terra: $1 input, $5 output per million tokens
- GPT-6.1 Sol: $2 input, $10 output per million tokens
- GPT-5.6 Terra: roughly double the new Terra price
The saving does not stop at the API. Inside ChatGPT Work and Codex, your plan goes further — OpenAI says one plan gets more done on Terra than on Sol or Astra — and small daily jobs make it pile up.
How Fast and How Good Is It?
One honest note before the numbers: most come from OpenAI’s own testing.
On Agents’ Last Exam, Terra beat GPT-5.6 Terra by 15 points at medium effort — and did it at roughly one-third of the cost per task.
Speed is the bold claim. OpenAI says Terra finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. But those timings are OpenAI’s own, not an outside lab’s, so hold the claim lightly until others check.
The coding numbers are hard to argue with. Terminal-Bench 4.0, max effort: Terra solved 50.3% of the tasks. The old Terra managed 21.5%. Better than double.
Mistakes fell sharply too. At low effort the error rate drops from 53.6% with the old Terra to 16.7% with the new one, and at max effort it falls again, to 4.3% — nearly level with Astra, the flagship, at a mid-tier price.
Outside voices back the story, mostly on price. Artificial Analysis called it the first Terra-class model since GPT-5.6 — more intelligent, at less than a third of the cost. GitHub’s early testing found it solving more coding tasks than the old Terra.
One last comparison, worth a slow read. Same test, max effort: Terra scored 60.0% at $1.01 per task. Claude Opus 5.5 scored 63.2% — higher by a whisker, but its bill was $8.82 per task, close to nine times Terra’s cost. So “which scored more?” is only half the question. The other half: “what did each point cost you?” On that one, Terra walks away with it.
Where You Can Use GPT-6.1 Terra
You can reach Terra in three places:
- ChatGPT Work — for Plus, Pro, Enterprise, and Edu users.
- Codex — the home for coding work. Just know this: Codex CLI users need version 0.159.0 or newer.
- The OpenAI API — for developers, under the name gpt-6.1-terra.
Launch-day coverage also has it on Amazon Bedrock and GitHub Copilot from day one. And the context window is seriously big — 1.05 million tokens in, 128,000 out. Feed it a very long document and it works through the whole thing in one go.
Why It Matters for Students
Let us be straight about student life. A huge slice of it is the same tasks, over and over.
Admission forms. Notes that look like a war zone. Key facts hiding inside a 40-page PDF. A rough draft to make neat. None of it is hard. All of it steals time.
So what shifts when the tool gets cheaper and quicker? Using it often stops feeling like a luxury — you hand it the small jobs too: main points from a long paper, a form filled from your notes, mistakes caught in your draft. Each job costs less, so each becomes worth doing.
One thing does not change. Check the work yourself — these tools still get things wrong, and for class work your own eyes are the final check. Always.
Why It Matters for Freelancers
A freelancer’s product is time. Hours are what you sell — and nothing chews through them like repeat computer work.
Client onboarding forms. Invoice entry. Files to rename. Data moving from this app into that one. Badly paid work, or unpaid — and it all stands in front of the real work.
Terra is aimed squarely at that pile. A stack of invoices goes in, clean records come back. Rough notes become a proposal draft. A short clip gets its trim.
For freelancers the price cut may matter even more than the speed. Repetition is why: ten jobs a day, five days a week, and the saving stacks until last month’s money-losing workflow pays for itself. Quicker turnaround means more clients in the same day.
Practical advice, then. Pick one safe, boring task. Check the result with hard eyes, and widen the list only as trust grows. Do not hand it full access to client money or private accounts on day one.
What to Watch Out For
Three cautions, worth taking seriously.
First, the testing. Most scores above are OpenAI’s own numbers, outside testing is still early, and your results may look nothing like lab results.
Second, computer use means the AI works inside your apps, touching your files. Powerful — and exactly why limits are not optional. Keep your own approval on for anything that moves money or speaks to a client.
Third, remember where Terra sits. Strong, yes — but not the top model, and for genuinely hard reasoning Sol or Astra may still do better. Terra’s ground is everyday jobs, won on price and speed.
OpenAI’s agent push did not start here. See our earlier post on OpenAI Dots AI assistant launch for how this agent race began.
The Bottom Line
So, plainly put, what is GPT-6.1 Terra? Not a showy release. A useful one.
It does daily computer work quicker than the model before it, costs half as much, and is live right now in ChatGPT Work, Codex, and the API.
Students and freelancers should feel it most. More jobs finished, less money out the door, and more of your day left for real clients, real study, real thought.
Give it one small task this week. Look at what comes back. If it hands you twenty minutes of your life back, it has already paid for itself.
Sources
OpenAI launch post, “OpenAI Ships GPT-6.1 Terra With Computer Use at Scale,” October 26, 2026 (as reported by AI Weekly). BleepingComputer, “OpenAI launches GPT-6.1 Terra with 3x faster AI agents, 50% lower pricing,” October 26, 2026. XDA Developers, “GPT-6.1 Terra brings more advanced computer agent use at a lower cost than Claude,” October 26, 2026.