Digiopedia

Image: Openai

OpenAI has introduced GPT-6.1 Terra, an updated AI model designed to improve visual reasoning, computer use, coding and everyday knowledge-work tasks. Announced on October 26, 2026, the model builds on GPT-5.6 Terra with improvements in task execution, factual accuracy and efficiency, while also reducing API costs.

Unlike AI models primarily designed to respond to questions or generate text, GPT-6.1 Terra is intended to handle more practical, computer-based work. It can interpret information displayed on a screen, interact with applications and complete tasks involving multiple steps. Examples include filling out forms, working with documents, interpreting technical drawings and editing video.

The release is part of OpenAI's broader GPT-6 model family, which includes models designed for different levels of capability, speed and cost. Terra focuses on providing a balance between performance and efficiency, making it particularly relevant to developers and businesses building AI-powered automation systems.

GPT-6.1 Terra Focuses on Computer Use and Visual Reasoning

One of the central improvements in GPT-6.1 Terra is its ability to understand visual information and translate that understanding into actions.

Computer-use systems need to do more than recognise text or describe an image. They must understand application interfaces, identify relevant controls, interpret information displayed on the screen and perform the appropriate actions in sequence.

For example, an AI agent handling an invoice may need to read the document, identify the supplier and payment information, open accounting software, locate the correct fields and enter the extracted details.

GPT-6.1 Terra is designed to handle this type of workflow. OpenAI highlights tasks such as filling out forms, interpreting technical drawings, editing videos and working with business applications as examples of its intended capabilities.

These improvements are particularly relevant to AI agents, which combine language models with tools and application access to carry out tasks beyond ordinary conversation.

OpenAI reports that GPT-6.1 Terra completes certain computer-use tasks more quickly than its predecessor and other models in its comparisons. At maximum reasoning effort, the company says Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol in its Agents' Last Exam evaluation.

Improved Performance on Computer-Use Benchmarks

OpenAI has used several benchmarks to demonstrate the improvements in GPT-6.1 Terra.

One of them is Agents' Last Exam, an evaluation of AI performance on complex professional tasks. These include workflows that require models to understand visual information, use tools and complete multiple stages of work.

At medium reasoning effort, GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra, according to OpenAI. The company also reports that the newer model achieves this result at approximately one-third of the cost per task.

Another evaluation is OSWorld 2.1, which measures how effectively AI systems interact with desktop applications to complete multi-step tasks.

These evaluations are important because computer-use systems face challenges that are different from conventional question-answering. An agent must understand where it is in an application, recognise the consequences of an action and recover when something does not happen as expected.

The results indicate that OpenAI is working to make computer-use capabilities more practical for applications involving repetitive business processes and desktop workflows.

However, benchmark performance does not guarantee that an AI agent will complete every real-world task correctly. Results can vary depending on the application, task complexity, available tools and operating environment.

Better Visual Understanding and Image Reasoning

Visual reasoning is another area receiving attention in GPT-6.1 Terra.

The model is designed to interpret visual information rather than relying exclusively on written instructions. This is useful for tasks involving diagrams, technical drawings, documents, application interfaces and other image-based material.

OpenAI evaluated the model using MMMU-Pro, a benchmark designed to assess multimodal understanding and reasoning across different subject areas.

According to the company's results, GPT-6.1 Terra outperforms GPT-5.6 Terra across every tested reasoning setting. It also approaches the performance of more capable models such as GPT-6 Astra and GPT-6.1 Sol, while operating at a lower cost.

This combination of visual understanding and computer interaction is significant for agent-based workflows. A model that can interpret a visual interface and act on the information it identifies can potentially handle more complete tasks instead of simply describing what appears on a screen.

GPT-6.1 Terra Improves Coding Performance

Although computer use and visual reasoning are major parts of the release, OpenAI has also reported improvements in software engineering.

The company evaluated the model on Terminal-Bench 4.0, which tests how well AI systems solve command-line tasks. These can include debugging, modifying software and resolving problems in development environments.

At maximum reasoning effort, GPT-6.1 Terra achieved a score of 50.3%, compared with 21.5% for GPT-5.6 Terra in the published results.

This represents a substantial improvement over its predecessor.

OpenAI also evaluated the model on Terminal-Bench-Science 0.1, which examines how AI systems use code to carry out scientific research tasks.

The results suggest that GPT-6.1 Terra is better suited to a wider range of technical workflows than the previous version. However, OpenAI still recommends GPT-6.1 Sol and GPT-6 Astra for the most complex software-engineering tasks.

The distinction is important: Terra is designed to offer an efficient balance of capability and cost rather than replace the more capable models across every demanding workload.

Fewer Factual Errors and Improved Reliability

OpenAI has also highlighted improvements in factual accuracy.

AI models can sometimes produce incorrect information while presenting it confidently. These errors, commonly referred to as hallucinations, remain an important challenge when models are used for research, business operations and other tasks where accuracy matters.

According to OpenAI's internal evaluation using difficult, error-inducing prompts, GPT-6.1 Terra reduced the factual error rate at low reasoning effort from 53.6% with GPT-5.6 Terra to 16.7%.

At maximum reasoning effort, the error rate fell by 18 percentage points to 4.3%.

These figures come from OpenAI's own internal benchmark and should not be interpreted as universal error rates across all questions or real-world applications.

Nevertheless, the reported improvement is relevant for developers and organisations that use AI systems to process information, prepare documents or assist with technical work.

Lower API Prices Compared With GPT-5.6 Terra

Cost is another major part of the GPT-6.1 Terra release.

OpenAI has reduced the model's standard API pricing by 50% compared with GPT-5.6 Terra. This is intended to make AI-powered automation more economical, particularly for applications that process large volumes of information or perform repetitive tasks.

The standard API pricing is:

Token category	GPT-6.1 Terra price
Input tokens	$1 per million
Cached input tokens	$0.05 per million
Output tokens	$5 per million

For developers, the reduction in pricing could make a difference when building applications that repeatedly analyse documents, process invoices, interact with software or perform other automated workflows.

Lower token costs can also make it more practical to run AI agents through multiple steps, although the total cost of an automated task depends on token usage, tool calls and other infrastructure expenses.

Availability in ChatGPT, Codex and the API

OpenAI began making GPT-6.1 Terra available on October 26, 2026.

The model is available to Plus, Pro, Enterprise and Edu users through ChatGPT Work and Codex. It is also accessible to developers through the OpenAI API under the model identifier gpt-6.1-terra.

An important distinction is that GPT-6.1 Terra is available in ChatGPT Work and Codex, rather than the standard ChatGPT chat experience.

API access is billed separately from ChatGPT subscriptions. Developers can integrate the model into applications and workflows using supported API functionality.

The model supports computer-use tools, function calling, structured outputs and other capabilities available through the Responses API.

GPT-6.1 Terra Specifications and Key Details

Category	Details
Model	GPT-6.1 Terra
Developer	OpenAI
Announcement date	October 26, 2026
Primary focus	Visual reasoning, computer use and writing
Previous version	GPT-5.6 Terra
Computer-use support	Yes
Visual input	Images and text
Context window	1,050,000 tokens
Maximum output	128,000 tokens
Reasoning settings	None, Low, Medium, High, Xhigh and Max
API model ID	gpt-6.1-terra
Input pricing	$1 per million tokens
Cached input pricing	$0.05 per million tokens
Output pricing	$5 per million tokens
ChatGPT availability	Work
Other availability	Codex and OpenAI API

What GPT-6.1 Terra Means for AI Agents

The release of GPT-6.1 Terra reflects a broader shift in how AI models are being developed and used.

Early generations of AI assistants were primarily designed to answer questions, summarise information and generate content. Newer systems are increasingly being developed to interact with software, understand visual interfaces and complete tasks involving several applications.

This approach makes AI agents useful for more than generating responses. They can potentially help automate administrative work, process documents, organise information and assist with technical workflows.

GPT-6.1 Terra is designed to make these capabilities more accessible by combining visual reasoning, computer interaction and improved efficiency at a lower API price.

It does not eliminate the need for human supervision, particularly when an automated action involves sensitive information, financial records or important business decisions. The reliability of an agent also depends on the tools and permissions it receives.

Still, the improvements OpenAI reports show how the company is developing models that can move beyond generating text and become more capable participants in everyday computer-based work.

For developers and businesses interested in building AI-powered automation, GPT-6.1 Terra represents another step towards making computer-using AI agents more practical to deploy at scale.