OpenAI launched GPT-6.1 Terra on October 26, a model designed to combine visual reasoning with precise actions on computers. The company presents it as a lower-cost option for everyday work that requires interpreting a screen and acting on it: completing forms, processing invoices, reading technical drawings, editing video, or assisting with software-development tasks. The model is now available in ChatGPT Work, Codex, and the API under the identifier `gpt-6.1-terra`. [1]

The proposition is not simply faster text generation. Terra is built to observe interfaces and carry out steps within digital workflows. OpenAI says the model improves substantially over GPT-5.6 Terra on professional computer-use tasks. On Agents’ Last Exam, an evaluation that includes turning architectural drawings into three-dimensional models and reviewing audiovisual material shot by shot, the company reports a 15-percentage-point advantage at medium reasoning effort and about one-third of the cost. This is a provider-run measurement and does not replace independent testing in production environments. [1]

Price is one of the launch’s central arguments. According to OpenAI, standard API rates are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol. The company argues that the reduction makes high-volume automation more practical, including extracting information from incoming invoices and entering it into accounting systems. Plus, Pro, Enterprise, and Edu users can access the model in ChatGPT Work and Codex; the same usage allowance is intended to cover more tasks than it would with higher-cost models. [1]

An independent publication summarizing the launch also described Terra as a faster and less expensive model for computer-based tasks, visual reasoning, and software development. That corroboration confirms the product’s existence and broad focus, but the detailed comparative figures still come from OpenAI. [3]

The release is accompanied by an addendum to the GPT-6 Astra system card. Published on October 26 as well, the document collects evaluations covering prompt injection, robustness, hallucinations, health, agentic behavior, and cybersecurity. In the cybersecurity safety table included in the addendum, GPT-6.1 Terra posts results comparable to recent GPT-6-family releases. OpenAI cautions that these measurements reflect model responses without all of the safeguards used in the final production system, an important distinction when interpreting the numbers. [2]

The company further says Terra is more likely than GPT-5.6 Terra to remain within the requested scope and to disclose limitations or blockers honestly. It also reports resistance to malicious instructions designed to divert a task. Those are relevant signals for a model that can operate interfaces, but they remain internal evaluations. Real-world safety will depend on the permissions, confirmations, logs, action limits, and human oversight surrounding the model. [1][2]

Terra extends a clear industry direction: moving agent and computer-use capabilities into models that are inexpensive enough for repetitive workloads. Its immediate impact will depend less on a single demonstration than on sustained reliability across real applications, ambiguous documents, and interface errors. The launch is material because it combines broad availability, product and API integration, and a price reduction that may change which automations are economically viable. Even so, its performance and safety advantages will need validation beyond the provider’s own tests.

Sources

[1] Read More — Introducing GPT-6.1 Terra — OpenAI
[2] Read More — Addendum to GPT-6 Astra System Card: GPT-6.1 Terra — OpenAI
[3] Read More — Tech News Today, October 27, 2026 — TechStartups