OpenAI introduced GPT-6.1 Terra on October 26, 2026, the successor to GPT-5.6 Terra. The fast model is meant to take over mainly office work on the computer: drafting quotes, analyzing technical drawings, filling out forms, and editing videos. Terra is being rolled out in ChatGPT Work and Codex for the Plus, Pro, Business, Enterprise, and Edu plans.
OpenAI does not mention an exception for Germany or the EU. The rollout is happening gradually, so Terra will not appear in all accounts at the same time.
Where do you find Terra in ChatGPT?
Terra is only available in Work mode, the agent for longer, multi-step tasks, and in Codex. OpenAI does not offer the model in the regular chat, and the Free and Go plans do not get Terra at all. In Work, you select GPT-6.1 Terra from the model list.
The usage limits of your plan stay the same. However, OpenAI promises that the same allowance covers more tasks than with GPT-5.6 Terra and significantly more than with GPT-6.1 Sol or GPT-6 Astra. For some Plus and Pro accounts, the default setting also changes: in the Work defaults, it is now Terra with medium reasoning effort instead of Sol with low effort. The reasoning effort determines how thoroughly a model thinks before answering.
What do OpenAI's own tests show?
All measurements come from OpenAI. For questions designed to provoke errors, the error rate with low reasoning effort dropped from 53.6 percent for GPT-5.6 Terra to 16.7 percent, and to 4.3 percent with maximum effort. In the programming test Terminal-Bench 4.0, Terra solved 50.3 percent of the tasks, while GPT-5.6 Terra reached 21.5 percent.
GPT-6.1 Terra in OpenAI's tests
Measurements from OpenAI at launch on October 26, 2026, compared with GPT-5.6 Terra
- 16.7 %Error rate on trick questions with low reasoning effort, previously 53.6 percent
- 4.3 %Error rate on trick questions with maximum reasoning effort
- 50.3 %tasks solved in the programming test Terminal-Bench 4.0, previously 21.5 percent
- +15percentage points in the computer test Agents’ Last Exam with medium reasoning effort
All values come from OpenAI's own tests
Source: OpenAI
For work on the computer, OpenAI cites the test Agents’ Last Exam, with tasks such as converting an architectural drawing into a 3D model. At medium effort, Terra scored 15 percentage points higher there than GPT-5.6 Terra, at about a third of the cost, and reached the level of GPT-6 Astra, which costs ten times as much per token. For the most difficult programming tasks, OpenAI continues to recommend GPT-6.1 Sol or GPT-6 Astra.
What does GPT-6.1 Terra cost?
In the ChatGPT subscription, OpenAI charges no extra fee; Terra counts toward the same allowance for Work and Codex. In the API, the model is called gpt-6.1-terra. The input there costs half as much as for GPT-5.6 Terra, and the output 58 percent less; compared with GPT-6.1 Sol, both prices are halved. The price reduction becomes noticeable with self-built automations, such as a workflow that reads incoming invoices and transfers them into accounting.
In comparison with Claude Opus 5.5, OpenAI puts the cost per task at maximum effort at around one ninth.
What this means for you
- Check the default setting: On your next Work task, quickly check which model and which effort level are set. For quotes and forms, which OpenAI itself names as a use case, Terra is sufficient, and your allowance lasts longer.
- Keep double-checking numbers: On the trick questions from OpenAI's own test, Terra with low effort got it wrong in every sixth case. For research involving names, data, and prices, it is best to set the effort higher and check the sources yourself.
- Don't switch plans too quickly: With Free or Go, Terra changes nothing about your chat. Switching to Plus is only worthwhile if you regularly need Work for multi-step tasks. How to use ChatGPT purposefully for texts and research is shown step by step in the ChatGPT Masterclass.
- Stick with Sol for large code projects: In Codex, you use Terra for small fixes; as OpenAI itself recommends, you are better off leaving complex rebuilds to Sol or Astra.
OpenAI introduced GPT-6.1 Terra on October 26, 2026.
Sources
