When to Use GPT-6.1 Terra

What OpenAI shipped on October 26

OpenAI introduced GPT-6.1 Terra as a fast model for computer use and everyday writing: drafting, reading technical drawings, filling forms, and other workday tasks on a screen. The API name is gpt-6.1-terra. OpenAI says it is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers on the API.

Standard API prices on that post are $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. OpenAI says those standard prices are half the list price of GPT-5.6 Terra and of GPT-6.1 Sol. The same post says a plan's usage allowance covers more tasks with Terra than with GPT-5.6 Terra, and substantially more than with GPT-6.1 Sol or GPT-6 Astra.

These are OpenAI's published prices and availability notes. They are not a CuratedAI measurement of speed, quality, or the bill you will actually see.

Read the computer-use claims as the lab's own scores

OpenAI's case for Terra is computer use at a lower token price. On its introduction post, the company reports these comparisons. None of them is an independent retest.

• Agents' Last Exam. At medium reasoning effort, OpenAI says Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, and performs on par with GPT-6 Astra at about one-tenth the price per token. At maximum effort, OpenAI says Terra is over three times as token-efficient as Claude Opus 5.5 and one-fourth its price per token, which the company puts at about nine times lower cost per task. OpenAI also says Terra finishes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol.
• OSWorld 2.1. At medium effort, OpenAI says Terra beats GPT-5.6 Terra's best score at a fraction of the cost per task. The example task on the post is building a 3D model from a technical drawing.
• MMMU-Pro. OpenAI says Terra beats GPT-5.6 Terra at every reasoning setting, and nearly matches GPT-6 Astra and GPT-6.1 Sol at a lower cost.
• Terminal-Bench 4.0. At maximum effort, OpenAI says Terra solves 50.3 percent of tasks, against a published 21.5 percent for GPT-5.6 Terra. The same paragraph says GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding tasks.
• Factual errors. On an internal set of error-inducing prompts, OpenAI says the error rate at low effort fell from 53.6 percent on GPT-5.6 Terra to 16.7 percent, and at maximum effort fell by 18 percentage points to 4.3 percent.

A company score can be useful as a routing hint. It does not tell you whether Terra will finish your invoice form, your internal app, or your repo. The system card is where OpenAI puts the longer safety write-up, including its Preparedness Framework rating. Treat that as the lab's classification, not as a test you can skip.

Match the model to the job in front of you

The price cut is the reason to try Terra. The reason to stay on a larger model is a task OpenAI itself still assigns upward. Use one real job, with a result you can check, before you move a queue.

Check a sample before you reroute the queue

• Name the done state. A filled field, a saved file, or a draft with the sources you supplied. "It used the computer" is not a result.
• Run the same ten tasks on Terra and on the model you use now. Count finishes, wrong clicks, and minutes. OpenAI's "three times faster" is its own timing, on its own exam.
• Price the whole task. A cheaper token can still lose if the model retries, asks for a higher reasoning effort, or hands the job back. Cached input at $0.05 only helps when the same prefix is actually cached.
• Keep the tools narrow. Computer use is the ability to act on a desktop. Give the trial the apps that job needs, and leave production accounts out of the first pass.
• Read the misses. If Terra fails the same step on most of the sample, that step still belongs on Sol, Astra, or a person. Do not average it away.

Where this release is easy to overread

"On par with Astra" on OpenAI's computer-use exam is not the same sentence as "use Terra for every GPT-6 job." The post separates those claims: visual and computer-use tasks are where Terra is meant to close the gap, and the hardest coding stays with the larger models. GPT-5.6 Terra's older directory notes, including its July prices, describe that earlier model. They are not the October 26 rate card.

ChatGPT Work and Codex are the surfaces OpenAI named. The post does not say Terra replaced every other model in ordinary ChatGPT. Check the picker in the product you actually have before you tell a team the default changed.

What to do this week

Pick one repeated computer-use or writing job. Run a ten-task sample on gpt-6.1-terra, record finishes and the token bill, and keep the larger model on anything the sample gets wrong. For the rest of the day's reporting, read the October 26 briefing. For the neighboring models already in the directory, see GPT-6.1 Sol and GPT-5.6 Terra. Reviewed October 27, 2026, from OpenAI's introduction post.