On October 26, 2026, OpenAI released a new AI model, GPT-6.1 Terra. Built with an emphasis on speed and low cost, it succeeds the earlier GPT-5.6 Terra. OpenAI says it is strong at operating computer apps, reading drawings and images in fine detail, and writing clear text. OpenAI's official page is also available in Japanese.
What is GPT-6.1 Terra?
OpenAI's current lineup includes GPT-6 Astra, its most capable model, and GPT-6.1 Sol, which is suited to tasks such as advanced coding. Terra sits in the lineup as the model that balances speed, performance and price.
OpenAI says it has in mind the kinds of tasks that come up in everyday work, such as drafting proposals, reading technical drawings, filling in forms and editing videos. One goal is to make it easier to automate high-volume work, such as reading incoming invoices and entering the line items into accounting software.
Operating computers and reading images
OpenAI measured the model's ability to operate a computer on Agents' Last Exam, a test made up of tasks drawn from professional work. The tasks include building a 3D model from architectural drawings and reviewing an animated film scene by scene to produce notes for a restoration team.
- With reasoning set to "medium," it scored 15 points higher than GPT-5.6 Terra at about one-third of the cost
- Under the same conditions, it matched the performance of GPT-6 Astra, the most capable model, at one-tenth of the price per token
- With reasoning set to the maximum, its cost per task was about one-ninth that of Anthropic's Claude Opus 5.5
- It finished tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol
On MMMU-Pro, which measures the ability to reason about images, OpenAI says it beat GPT-5.6 Terra at every setting and came close to the scores of GPT-6 Astra and GPT-6.1 Sol at a much lower cost. On OSWorld 2.1, which measures the ability to carry out multi-step tasks in desktop apps, it also topped GPT-5.6 Terra's best score.
Fewer errors and better coding
This year OpenAI has focused on reducing "hallucinations," in which an AI gives plausible but wrong answers. On an internal test made up of questions designed to lure the model into errors, the error rate with low reasoning fell from 53.6% for GPT-5.6 Terra to 16.7%. With maximum reasoning it was 4.3%, roughly the same level as GPT-6 Astra and GPT-6.1 Sol.
In coding, it solved 50.3% of the problems on Terminal-Bench 4.0, which measures the ability to solve problems from the command line, with maximum reasoning. That is more than double GPT-5.6 Terra's 21.5%. However, OpenAI says GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding.
All of these figures come from OpenAI's own announcement and have yet to be independently verified.
Pricing and availability
GPT-6.1 Terra is available from the day of the announcement in the following places.
- ChatGPT Work and Codex: all Plus, Pro, Enterprise and Edu users
- The OpenAI API: model name gpt-6.1-terra
Standard API pricing per 1 million tokens is 1 dollar for input, 0.05 dollars for cached input and 5 dollars for output. That is 50% lower than the standard pricing of GPT-5.6 Terra and GPT-6.1 Sol. In ChatGPT Work and Codex, OpenAI says the same usage allowance covers more work than with GPT-5.6 Terra, and the gap is even larger compared with GPT-6.1 Sol and GPT-6 Astra.
How safety is handled
OpenAI says GPT-6.1 Terra has reached "Critical," the highest level in its internal Preparedness Framework safety standards, in the area of cybersecurity. As a result, it uses the same safeguards OpenAI built for GPT-6 Astra.
OpenAI also says the model is more likely than GPT-5.6 Terra to stay within the scope of the task it was given and to tell users honestly about anything that might get in the way of the work. It is also said to be more resistant to malicious instructions that try to divert it from its task (prompt injection). Details are published in the system card.
CoAI's take
GPT-6.1 Terra delivers performance close to the top-tier models at half the price and several times the speed. For companies that want to hand everyday office tasks to AI, and for developers running large volumes of work through the API, it looks like an option worth trying first.
On the other hand, within ChatGPT it is available only in the work-focused ChatGPT Work and in Codex, and in both cases only on the Plus, Pro, Enterprise and Edu plans. The announcement does not mention the free plan or the Business plan. The statement that it reached the highest level in the cyber area is also a reminder that AI capable of operating computers needs careful handling. When using it with in-house systems, it is best to define which operations it is allowed to perform and to introduce it alongside a process in which people check the results.
