OpenAI has announced GPT-6.1 Terra, a new AI model built to get work done faster and at lower cost. Beyond writing content or analysing images, Terra can understand software interfaces and carry out actions such as data entry, filling forms or processing documents. It is a notable step in the race to turn AI from a chat tool into an assistant that does real work.
What makes GPT-6.1 Terra special?
On 26 October 2026, OpenAI officially introduced GPT-6.1 Terra, an upgrade from GPT-5.6 Terra. Unlike models focused only on reasoning or text generation, GPT-6.1 Terra is optimised for three groups of tasks: image understanding, computer use and natural writing.
According to OpenAI, the model can help read technical drawings, recognise information in invoices, fill forms, edit content and perform tasks across app interfaces. For example, instead of a worker opening each invoice, reading the amount and typing it into accounting software, a system using Terra can be designed to recognise the data and carry out the entry steps. Still, real-world capability depends on the tools it is granted, system configuration and a company's review processes.
Three times faster, significantly cheaper
The standout of GPT-6.1 Terra is the balance between speed, capability and cost. By OpenAI's own results, on the Agents' Last Exam benchmark at medium reasoning effort, Terra scores about 15 percentage points higher than GPT-5.6 Terra at roughly one-third the cost.
Notably, Terra can match GPT-6 Astra on this benchmark while its token cost is only a fraction. OpenAI also says Terra completes the benchmark tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. These are results on specific OpenAI-reported measurements and do not mean Terra is three times faster in every real-world case.
API pricing from just $1 per million input tokens
OpenAI announced the standard pricing of GPT-6.1 Terra as follows:
- $1 per million input tokens.
- $0.05 per million cached input tokens.
- $5 per million output tokens.
By the company's figures, this standard pricing is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. The lower cost matters most for businesses handling large volumes of repetitive work, such as sorting documents, checking information or processing customer records. However, real running costs may still include tool, infrastructure and extra review charges. For requests above about 272,000 input tokens, long-context pricing is higher.
Fewer factual errors, better coding
OpenAI also highlighted accuracy gains. In the company's internal test using prompts designed to elicit wrong answers, Terra's error rate at low reasoning effort is reported to fall from 53.6% to 16.7%; at maximum reasoning effort, it drops to 4.3%.
On coding, GPT-6.1 Terra scores 50.3% on Terminal-Bench 4.0 at maximum reasoning effort, well above the 21.5% reported for the previous generation. Even so, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra for the hardest coding tasks and deep research. These numbers show improvement in an evaluation setting, but do not guarantee the model will not make mistakes in real use.
Who can use GPT-6.1 Terra?
GPT-6.1 Terra is rolling out in ChatGPT Work and Codex for eligible paying users. Developers can also access the model through the OpenAI API as gpt-6.1-terra.
One thing to note: Terra does not appear as a direct option in the regular ChatGPT Chat interface. Users reach it through supporting experiences such as Work, Codex or API integration. Availability on each account may depend on the plan, region and rollout schedule.
AI is shifting from answering to doing the work
The arrival of GPT-6.1 Terra reflects a major trend in AI: the ability to handle real work is becoming as important as the ability to answer well. For a business, an AI model does not have to top every leaderboard to add value.
If it can read documents accurately, operate software reliably, finish tasks quickly and cost a reasonable amount, that model may fit daily operations better. That is also why OpenAI stressed computer use and cost efficiency at this launch.
GPT-6.1 Terra raises not only the question of how smart AI is, but a more practical one: how much work can AI finish on the same budget?
Key figures
| Item | Value |
|---|---|
| API input price | $1 / 1M tokens |
| API output price | $5 / 1M tokens |
| Standard price cut | ~50% vs GPT-5.6 Terra and GPT-6.1 Sol |
| Context window | ~1.05 million tokens |
Frequently asked questions
- How is GPT-6.1 Terra different from a normal model? — It is optimised to understand the screen and operate software (fill forms, process documents), not just answer questions like a chatbot.
- How much does GPT-6.1 Terra cost? — Per OpenAI: $1 per million input tokens, $0.05 for cached tokens and $5 per million output tokens — about 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
- Who can use GPT-6.1 Terra? — It is rolling out in ChatGPT Work and Codex for eligible paying users; developers call it via the API as gpt-6.1-terra.
