OpenAI unveils GPT-6.1 Terra – faster and half the price

OpenAI has unveiled GPT-6.1 Terra, a new AI model for working with apps, visual data, and code. It replaces GPT-5.6 Terra and costs half as much under standard API pricing, while coming close to the more expensive GPT-6.1 Sol and GPT-6 Astra in several tests.

Working with interfaces and images

GPT-6.1 Terra is designed primarily for tasks that require more than just generating text. It can analyze images and technical drawings, interact with app interfaces, fill out forms, and perform multistep actions on a computer. Examples cited by OpenAI also include video editing and creating 3D models from drawings.

To assess the new model’s capabilities, OpenAI used several benchmarks. The results show a marked improvement over GPT-5.6 Terra, particularly in coding and working with desktop apps.

*The difference on Agents’ Last Exam was measured at the medium reasoning level. Terminal-Bench 4.0 results are shown at the maximum level. Higher success rates are better. For factual errors, a lower rate indicates greater accuracy.

In addition to improved accuracy, GPT-6.1 Terra requires less computing power. On Agents’ Last Exam, completing a task costs about three times less than with its predecessor. At the maximum reasoning level, the model also uses more than three times fewer tokens than Claude Opus 5.5, while the total cost of solving a task is, according to OpenAI, about nine times lower.

The company says GPT-6.1 Terra completes similar tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.

Fewer errors and better coding

One notable improvement is factual accuracy. On an internal set of prompts specifically designed to provoke errors, GPT-6.1 Terra significantly outperformed its predecessor. At the maximum reasoning level, its result came close to those of GPT-6 Astra and GPT-6.1 Sol.

However, these figures are from a specific internal OpenAI test and do not reflect the frequency of errors in everyday user conversations.

There has also been substantial progress in coding. Terminal-Bench 4.0 tests AI agents’ ability to solve tasks independently through the command line, including fixing errors in code. As the table shows, GPT-6.1 Terra’s score more than doubled.

On Terminal-Bench-Science 0.1, which focuses on using code for scientific tasks, Terra outperformed GPT-6.1 Sol and came close to Claude Opus 5.5. OpenAI estimates that the cost of completing these tasks was more than five times lower than with Anthropic’s model.

However, GPT-6.1 Terra does not replace the more advanced models in every scenario. For the most challenging development tasks, OpenAI still recommends GPT-6.1 Sol and GPT-6 Astra.

Safety

According to internal tests, GPT-6.1 Terra is more likely than GPT-5.6 Terra to stay on task and tell users about limitations or obstacles that prevent it from completing a task.

OpenAI also says the model is more resistant to prompt injection attacks, in which malicious instructions in a webpage, document, or other external content attempt to make an AI agent ignore the user’s original request.

Under the Preparedness Framework, the model reached the Critical level for cybersecurity capabilities. OpenAI is therefore applying to GPT-6.1 Terra the safeguards developed for GPT-6 Astra.

Availability and pricing

GPT-6.1 Terra became available on October 26, 2026, to ChatGPT Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Developers can access the model through the API using the identifier gpt-6.1-terra.

Standard API pricing is:

• $1 per 1 million input tokens;
• $0,05 per 1 million cached input tokens;
• $5 per 1 million output tokens.

That is 50% less than the standard price for GPT-5.6 Terra and GPT-6.1 Sol. In ChatGPT Work and Codex, OpenAI has also increased the amount of work users can do within existing limits: at the same limit, GPT-6.1 Terra should handle more tasks than GPT-5.6 Terra, GPT-6.1 Sol, and GPT-6 Astra.