GPT-6.1 Terra announced: Faster, lower-cost computer use

OpenAI announced GPT-6.1 Terra on 26 October 2026. Combining visual reasoning with the ability to use computer applications, the model is aimed at tasks such as interpreting architectural drawings and completing multi-step work in desktop software. GPT-6.1 Terra is available the same day in ChatGPT Work and Codex to Plus, Pro, Enterprise and Edu users, while developers can access it through the OpenAI API.

From architectural drawings to computer tasks

OpenAI says its Agents’ Last Exam test, which measures professional workflows, includes an example of creating a 3D model from architectural drawings. According to results shared by the company, at the medium reasoning setting GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra on this test, while the task cost was about one-third as much. OpenAI also says that, at the same setting, the model delivers performance comparable to GPT-6 Astra at one-tenth the price per token.

OSWorld 2.1 evaluates how agents handle multi-step tasks in desktop applications. Its examples include preparing a 3D model from a technical drawing. According to OpenAI, at medium reasoning GPT-6.1 Terra surpassed GPT-5.6 Terra’s best score at a lower cost per task. These results reflect performance on tests disclosed by the company; they do not, on their own, guarantee success with different types of drawings and software in real-world use.

Speed, accuracy and API pricing

OpenAI says GPT-6.1 Terra completes tasks more than three times faster than GPT-6.1 Sol and GPT-6 Astra. In the company’s comparison, the model’s standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol. Pricing is listed at $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens.

The company says its internally developed error-prone prompt test showed the error rate at the low reasoning setting fall from 53.6% for GPT-5.6 Terra to 16.7%. At the highest setting, the rate is reported as 4.3%. For coding, OpenAI says the model solved 50.3% of tasks on Terminal-Bench 4.0 at the highest reasoning setting. That is more than twice the GPT-5.6 Terra score published by OpenAI. These results are based on the company’s tests and should be validated separately in real projects.

Impact on architecture and visualization workflows

For architecture firms and visualization teams, the standout feature is the ability to handle technical drawings and computer-based actions within the same task flow. Teams evaluating repetitive desktop work or processes that turn drawings into models can test the model with their own files and applications. However, the announcement provides no details on compatibility with specific architectural software, hardware requirements or the quality of models generated from drawings.

How users access the model also depends on their workflow: GPT-6.1 Terra is available through ChatGPT Work and Codex on the specified subscription plans, while developers can opt for the API. Although per-token API pricing has been announced, total spending depends on usage. Firms should also review licensing terms, their own workloads and data-processing requirements.

Access and evaluation

OpenAI says GPT-6.1 Terra retains the safety protections developed for GPT-6 Astra and met the Critical threshold under the Preparedness Framework in its cybersecurity assessment. The model’s test results for computer use and visual interpretation are notable for the potential to automate architectural work. Still, it is important to measure speed, error rates and cost using real project files before moving to production.