Half the price of the model it replaces. More than three times faster than OpenAI’s top models. That’s the pitch for GPT-6.1 Terra, the new model OpenAI released on Monday, 26 October. It’s aimed at everyday desk work, like filling in forms, reading drawings and clicking around apps for you. So is it actually good, or just cheap? Let’s look at what we know.
So what is GPT-6.1 Terra, exactly?
OpenAI sorts its models into four classes. From cheapest to most capable, they’re Luna, Terra, Sol and Astra. Terra is the middle child. It’s not the smartest in the family, but it’s the one you’d send to do most of the actual work.
Here’s the odd bit. When OpenAI launched its GPT-6 Sol and Luna models in September, there was no new Terra. So this release fills that gap, and it replaces GPT-5.6 Terra.
OpenAI calls it “a fast, efficient model that combines detailed visual reasoning and precise computer use.” In plain words, computer use means the AI can look at your screen and click, type and scroll the way you would. Then it can finish a task across several apps without you steering every step.
According to 9to5Mac, OpenAI has jobs like writing proposals, reading engineering drawings, completing forms and even editing video in mind. The company also says the model writes more clearly and naturally than before.
You can use it now on ChatGPT Plus, Pro, Enterprise or Edu, inside ChatGPT Work and Codex. ChatGPT Work is OpenAI’s agent mode for jobs across apps, while Codex is its coding tool. Free users, for now, get GPT-5.6 Luna instead.
The price is the real headline
Honestly, the price is the story here. Developers pay $1 per million input tokens and $5 per million output tokens, according to Artificial Analysis. The old GPT-5.6 Terra cost $2 and $12. A token, by the way, is a small chunk of text, often just part of a word.
But OpenAI goes further. In its launch announcement, it says Terra matches its top model, GPT-6 Astra, at one-tenth the price per token. It also says Terra finishes tasks more than three times faster on average than both Astra and GPT-6.1 Sol.
Artificial Analysis ran its own numbers, too. It puts the cost per task at maximum effort at $0.40 for GPT-6.1 Terra, compared with $1.40 for GPT-5.6 Terra. Its verdict: “more intelligent for less than a third of the cost.”
Why does that matter? Because agents are hungry. An AI that clicks through your apps keeps reading screens and planning the next step, so it burns through tokens fast. Cut the price per token and suddenly a lot more tasks are worth handing over.
That’s also good news for your ChatGPT limits. OpenAI says your allowance in ChatGPT Work and Codex lasts longer with Terra than with the older model, and much longer than with Sol or Astra.
Does GPT-6.1 Terra get things right?
Cheap is great. But cheap and wrong isn’t. So here’s what the accuracy numbers say.
OpenAI ran an internal test with prompts designed to trip models up. At low reasoning effort, the error rate fell from 53.6% on GPT-5.6 Terra to 16.7% on the new model. At maximum effort, it dropped 18 points to 4.3%, which is close to Astra and Sol.
Coding improved, too. On Terminal-Bench 4.0, a test of coding jobs done in a command-line terminal, Terra solved 50.3% of tasks at maximum effort. The old Terra only managed 21.5%, so that’s more than double.
Now the catch. Every one of those figures comes from OpenAI, and the error test is its own, as Decrypt pointed out. So we looked for an outside view.
Artificial Analysis, which tests models independently, found a big drop in made-up answers. In its AA-Omniscience test, the hallucination rate fell to 38%, down from 88% for GPT-5.6 Terra. A hallucination, if you haven’t met the term, is when an AI states something false with total confidence. We’ve all seen plenty of those, especially from older models. On its Coding Agent Index, Terra scored 48, while GPT-6.1 Sol scored 54 and GPT-6 Luna scored 31.
Where it still falls short
Let’s not get carried away. OpenAI itself says that for in-depth research or complex coding, you should still reach for GPT-6.1 Sol or GPT-6 Astra. That’s a pretty clear signal about where Terra sits.
Also, look again at those numbers. Even on its best setting, the model got 4.3% of tricky questions wrong in OpenAI’s test. On the quick setting, it was 16.7%. And it still failed about half the coding tasks in Terminal-Bench. Decrypt summed it up with a cheeky headline, calling it OpenAI’s “Half-Good AI Model at Half the Price.”
That’s a bit harsh, in our view. But it’s a fair reminder. An agent that’s wrong about one time in six on a quick setting needs a human checking its work, especially for anything involving money or customers.
There’s one more thing to keep in mind. Artificial Analysis found that Terra, at maximum effort, still scores 4 points below GPT-6.1 Sol on its Intelligence Index. Then again, it’s 10 points above GPT-6 Luna. So it really is the middle option, just a much better-value one.
Okay, but should you care?
If you’re a regular ChatGPT user, the change is simple. Your paid plan gets a faster, more accurate model for agent tasks, and your limits go further. That’s nice, though not life-changing.
The bigger shift is for small teams. Think of a five-person startup in Yerevan or Gyumri. Until now, letting an AI agent fill in forms or tidy spreadsheets all day could get pricey fast. At $1 and $5 per million tokens, the maths starts to look a lot friendlier. And since the model works through screens and apps, you may not need a developer to wire everything up first.
We’ve written before about how ChatGPT’s Chrome extension can click around your tabs for you. GPT-6.1 Terra is the kind of engine that could make tools like that cheap enough to use daily. If you’re still fuzzy on the basics, our guide to what an AI agent is covers the idea in a few minutes.
So here’s our take. GPT-6.1 Terra won’t replace your smartest model for hard problems. But for the boring, repetitive clicking that eats your week, it might be the first one that’s cheap enough to hand the job to. Would you trust it with your inbox yet?