OpenAI launches GPT-6.1 Terra with 50% lower prices and higher accuracy

A cheaper path to automation

OpenAI has officially released GPT-6.1 Terra, a new model positioned as the company's everyday workhorse. This update arrives with a significant price cut for developers who use the model via standard APIs. Specifically, the new model costs 50% less than the previous GPT-5.6 Terra version.

The company designed Terra to handle routine office work rather than the most complex research problems. It is built to perform tasks like reading on-screen content, clicking through different applications, and drafting various types of replies. By lowering the cost, OpenAI aims to make these types of automated workflows more accessible to a wider range of users.

The model is now available to users with Plus, Pro, Enterprise, and Edu subscriptions. These users can access it through ChatGPT Work and Codex. Developers can also plug the model into their own custom applications using the gpt-6.1-terra identifier.

Improving accuracy and speed

One of the primary goals of this release is to reduce the frequency of mistakes made by the AI. In internal tests designed to trick the models into making errors, the rate of factual mistakes at low reasoning effort fell from 53.6% to 16.7%. At maximum effort, the error rate dropped by 18 points to reach 4.3%.

Speed is another major factor in the new model's performance profile. OpenAI reports that Terra finishes tasks more than three times faster on average than both the GPT-6 Astra and the GPT-6.1 Sol models. This makes it a strong candidate for high-volume tasks where quick turnaround times are necessary for productivity.

The model also showed significant improvements in technical environments. On the Terminal-Bench 4.0, which evaluates how well coding agents handle tasks in a computer's command-line terminal, Terra solved 50.3% of tasks at maximum effort. This is a notable jump from the 21.5% success rate seen in the older GPT-5.6 Terra model.

Mastering computer use

A key capability of the Terra model is what OpenAI calls 'computer use.' This refers to an AI's ability to interact with a computer interface in a way that mimics human behavior. The model can see what is happening on a screen and then perform actions like clicking buttons or typing text into fields.

This capability is particularly useful for navigating complex websites or moving data between different software applications. Instead of just generating text, the model can act as an agent to complete multi-step processes. This makes it a core component of ChatGPT Work, which is the agent mode designed to browse and complete tasks across various apps.

While Terra is the primary model for these everyday interactions, it is not intended to replace the most powerful models in the lineup. For users who need to perform in-depth research or handle extremely complex coding projects, OpenAI still recommends using the GPT-6.1 Sol or GPT-6 Astra models.

The history of the Terra model

The Terra model first appeared in a limited preview in June as part of the GPT-5.6 family. It was initially requested by the Donald Trump administration before being opened to the general public in July. At that time, it was already positioned as a model that could compete with higher-tier options at roughly half the cost.

Following that initial release, OpenAI introduced the GPT-6 Sol and Luna models in September. These were launched shortly after Anthropic released its Claude Opus 5.5 model, but they did not include a new version of Terra. The company chose to focus on those specific new models during that launch window.

The current release of GPT-6.1 Terra represents a refinement of the existing middle-tier offering. It provides a more polished experience for the general public while maintaining the cost-effectiveness that defined its earlier iterations. It serves as a bridge between the basic Luna model for free users and the high-end Sol and Astra models.