OpenAI unveils GPT-6.1 Terra
- Maxime Hiez
- OpenAI
- 31 Oct, 2026
Introduction
OpenAI released GPT-6.1 Terra on October 26, 2026, a fast, efficient model that combines detailed visual reasoning and precise computer use with clearer writing. According to OpenAI, it is a clear step up from GPT-5.6 Terra on everyday work tasks. It costs half the per-token price of GPT-6.1 Sol.
Check the article on GPT-6.1 Sol HERE.
Performance
All the results below are OpenAI’s own, obtained with its own methods :
| GPT-6.1 Terra | Baseline (GPT-5.6 Terra) | |
|---|---|---|
| Terminal-Bench 4.0 (maximum effort) | 50.3% | 21.5% |
| Factual error rate (low effort) | 16.7% | 53.6% |
The factual error rate is measured on an internal benchmark of prompts likely to induce mistakes. At maximum effort, it drops by 18 points to 4.3%, getting close to GPT-6 Astra and Sol.
On Agents’ Last Exam, a benchmark of professional tasks, Terra scores 15 percentage points higher than GPT-5.6 Terra at medium effort, at about a third of the cost, and OpenAI says it is comparable to Astra at a tenth of the price per token. At maximum effort, again according to OpenAI, it is more than three times as token-efficient as Claude Opus 5.5, at about a ninth of the cost per task. This comparison with Anthropic is OpenAI’s own and has not been verified. Terra beats GPT-5.6 Terra on MMMU-Pro (visual reasoning) and, at medium reasoning effort, on OSWorld 2.1 (computer use).
OpenAI adds that it completes tasks more than three times faster than Astra and Sol, but keeps those two models as the best choices for the most complex coding.
Safety and alignment
According to OpenAI, Terra reaches “the Critical threshold in cybersecurity” defined by its preparedness framework and keeps the protections developed for Astra. OpenAI also says that, in its evaluations, Terra stays within the scope of the task better than GPT-5.6 Terra, reports its limits more honestly, and showed strong resistance to prompt injection. The announcement gives no rate : these are OpenAI’s tests, not a guarantee. A system card details these points.
Vision, computer use and API
Highlighted use cases include writing proposals, interpreting technical drawings, filling in forms, and even video editing. The model accepts text and images as input and produces text. In the Responses API, it gives access to computer use, MCP, web and file search, the code interpreter and other tools.
It is exposed under the identifier gpt-6.1-terra, with a 1 050 000-token context window (922 000 input, 128 000 output) and a knowledge cutoff of June 1, 2026. The reasoning.effort parameter ranges from “none” to “max” (“medium” by default); unlike Astra and Sol, Terra accepts “none”. Tool calling requires the Responses API.
Pricing and availability
Terra has been available since October 26, 2026 to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Business is not mentioned. The same usage allowance covers more tasks there than with GPT-5.6 Terra, and noticeably more than with Sol or Astra, with no figure in the announcement. Developers can access it through the API as gpt-6.1-terra, at the following Standard rates, per million tokens (above 272K input tokens, the rate applies to the whole request) :
| Up to 272K | Above 272K | |
|---|---|---|
| Input | 1$ | 2$ |
| Output | 5$ | 7.50$ |
| Cache (read) | 0.05$ | 0.10$ |
| Cache (write) 30 min | 1.25$ | 2.50$ |
| Cache (write) 12 h | 2.50$ | 5$ |
Batch and Flex modes cost 50% less than Standard mode, regional processing adds 10%, and Fast mode is billed at double the rate.
On the pricing page, Sol is 2$ input and 10$ output, Astra 10$ and 50$, and GPT-5.6 Terra 2$ and 12$. Terra therefore costs exactly half as much as Sol and one tenth of Astra’s price. Against GPT-5.6 Terra, only input is half price : output drops from 12$ to 5$.
note
Conclusion
Terra targets high-volume workloads that read images and screens, where the per-token price matters more than reasoning depth. The figures from OpenAI, including the drop in error rate and the comparison with Opus 5.5, remain the vendor’s own : testing on your own workflows, at the same effort as GPT-5.6 Terra, will tell you whether Terra is enough before you move up to Sol.
Sources
OpenAI - Introducing GPT-6.1 Terra
Did you enjoy this post ? If you have any questions, comments or suggestions, please feel free to send me a message from the contact form.
Don’t forget to follow us and share this post.