OpenAI has launched GPT-6.1 Terra, a faster, cheaper model built for computer use, visual reasoning and everyday agent work. It costs $1 per million input tokens and $5 per million output tokens, half the price of GPT-6.1 Sol. If you run agents that click through apps all day, that price cut is the headline.
Here are the facts you need before reading further.
- Launched October 26, 2026 in ChatGPT Work, Codex and the API.
- Price: $1 input, $5 output, $0.05 cached input per million tokens.
- ChatGPT access: Plus, Pro, Enterprise and Edu plans.
- API model name:
gpt-6.1-terra, not available on the free API tier. - Context window: about 1.05 million tokens in the API, 500,000 in Codex.
Every performance figure below comes from OpenAI’s own announcement, as relayed by press coverage. Independent tests will likely tell a more nuanced story over the coming weeks.
What GPT-6.1 Terra is and where it fits
Terra is the budget-friendly tier in OpenAI’s lineup. It arrived about four weeks after GPT-6.1 Sol and upgrades the Terra tier that first appeared with GPT-5.6, according to iThinkDifferent.
OpenAI pitches it for three jobs: operating software on a screen, understanding images, and writing clearer answers. It is not meant to replace the heavier models for the hardest coding problems.
Think of it as the workhorse. You hand it the repetitive, high-volume jobs, such as checking dashboards, moving data between apps or reading screenshots, and save the pricier models for work where one wrong step is expensive.
GPT-6.1 Terra price vs Sol: the numbers
The pricing gap is the easiest thing to compare, so here it is side by side. Terra’s standard API rates sit 50 percent below both GPT-5.6 Terra and Sol.
| Item | GPT-6.1 Terra | GPT-6.1 Sol |
|---|---|---|
| Input (per million tokens) | $1 | $2 |
| Output (per million tokens) | $5 | $10 |
| Cached input (per million tokens) | $0.05 | $0.10 |
| Best suited for | Fast computer use, visual tasks, high-volume agents | The most complex coding work |
Cached input is where agents save the most. Long agent runs resend the same context again and again, so a $0.05 cache rate adds up quickly.
Computer use and speed: where Terra pulls ahead
OpenAI says Terra finishes computer-use tasks more than three times faster than both GPT-6 Astra and Sol. On its Agents’ Last Exam benchmark, the medium-effort setting scores 15 points higher than GPT-5.6 Terra at roughly one-third of the cost.
The company also claims it matches GPT-6 Astra on that test at one-tenth the price per token. On OSWorld 2.1, a multistep desktop benchmark, it matches the best medium-effort score of GPT-5.6 Terra for far less money.
This is the same race you can see in Cursor’s computer-use agents and in ChatGPT’s Chrome extension that controls your tabs. Cheaper screen-driving models make those tools affordable to run all day, and that is what Terra is really aimed at.
Speed matters as much as price here. An agent that takes three times longer per task also ties up your machine and your patience, so a faster model cuts costs twice.
Fewer wrong answers
OpenAI’s error-inducing prompt test shows the low-effort error rate dropping from 53.6% on GPT-5.6 Terra to 16.7%. At max effort, the error rate is 4.3%, close to Astra and Sol.
Where Sol and Astra still win
Terra is not the best choice for everything. OpenAI itself says Sol and Astra remain better for the most complex coding work.
Terra scores 50.3% on Terminal-Bench 4.0 at max effort, up from 21.5% for GPT-5.6 Terra. That is a big jump, yet it is still a step below the flagship tiers. On a science coding test, OpenAI says Terra lands within five points of Claude Opus 5.5 at under one-fifth of the estimated cost.
Security teams should also note that OpenAI rates Terra as meeting the Critical cybersecurity threshold under its Preparedness Framework. It keeps the protections that GPT-6 Astra has.
How to get GPT-6.1 Terra today
Access depends on how you work. Pick the route that matches you:
- In ChatGPT Work and Codex: Plus, Pro, Enterprise and Edu users can select it now, and its usage allowance stretches further than Sol’s.
- In the API: call
gpt-6.1-terrathrough the Responses API, Chat Completions or Batch. Tool calls need the Responses API, per Nerds Chalk.
Developers on the free API tier are out of luck for now. You need a paid tier to try it.
One practical tip: start with the medium effort setting. OpenAI’s strongest cost claims are tied to it, and you can raise the effort later only for tasks that fail.
Frequently Asked Questions
How much does GPT-6.1 Terra cost?
It costs $1 per million input tokens and $5 per million output tokens. Cached input is $0.05 per million tokens.
Is GPT-6.1 Terra better than GPT-6.1 Sol?
Not across the board. Terra is cheaper and faster for computer use and visual work, but OpenAI says Sol and Astra are still better for the most complex coding.
Who can use GPT-6.1 Terra in ChatGPT?
Plus, Pro, Enterprise and Edu subscribers can use it in ChatGPT Work and Codex. The rollout began on October 26, 2026.
What is the API name for GPT-6.1 Terra?
The model name is gpt-6.1-terra. The free API tier does not support it.
Should you switch to Terra?
If your agents spend most of their time clicking, reading screens and filling forms, test Terra first, because halving the bill with a faster model is hard to ignore. Keep Sol or Astra for tough coding jobs, and run your own tasks before trusting any vendor benchmark.
Leave a Reply