OpenAI launches GPT-6.1 Terra for cheaper office automation

A model for everyday office work

OpenAI officially released GPT-6.1 Terra on October 26. This new model is specifically built to handle the types of screen-based tasks that occupy a standard workday. It is designed to assist with drafting proposals, reading technical drawings, and filling out various forms.

The company is positioning Terra as a tool for high-volume automation. One specific example provided by OpenAI involves reading incoming invoices and automatically keying those details into accounting software. This suggests a focus on practical utility rather than just creative conversation.

The model is now available to paying ChatGPT users and through the API. It joins the existing lineup alongside GPT-6.1 Sol and GPT-6 Astra. OpenAI intends for Terra to be the primary choice for users who prioritize speed and lower costs over maximum reasoning power.

Performance and cost benchmarks

The pricing for GPT-6.1 Terra is significantly lower than many of its competitors. For developers using the API, the output costs $5 per million tokens and the input costs $1 per million. These costs drop even further to $0.05 per million when using cached data.

OpenAI claims that Terra offers a substantial discount on professional computer tasks. On a benchmark called Agents’ Last Exam, Terra scored 15 percentage points higher than the previous GPT-5.6 Terra while costing roughly one-third of the price. It also matches the performance of GPT-6 Astra at only a tenth of the price per token.

When pushed to maximum effort, the company says Terra is about nine times cheaper per task than Anthropic’s Claude Opus 5.5. These figures are provided by OpenAI to demonstrate how the model balances performance with affordability for large-scale operations.

Improving factual accuracy

One of the most important updates for business users is the improvement in factuality. OpenAI tested Terra on an internal set of prompts specifically designed to provoke errors. When the reasoning setting was kept low, Terra only got 16.7% of the prompts wrong.

This is a major improvement over the previous GPT-5.6 Terra, which had a 53.6% error rate on the same set of prompts. If a user chooses to set the reasoning to maximum, the error rate drops even further to just 4.3%.

While these improvements are significant, OpenAI is transparent about the model's limitations. For the most difficult coding projects, the company still recommends that users stick to the more powerful Astra or Sol models.

Security and availability

Because of its capabilities, Terra is subject to specific safety standards. Under OpenAI’s Preparedness Framework, the model reaches the Critical level for cybersecurity. This means it retains the same rigorous safeguards that were built into the more powerful Astra model.

The model is being rolled out to several different tiers of service. It is reaching ChatGPT Work and Codex for users on Edu, Enterprise, Pro, and Plus plans. Developers can access the model directly by calling it as gpt-6.1-terra.

The rollout ensures that both individual power users and large organizations can access the new tool. By providing a cheaper path to automation, OpenAI is making it easier for firms to integrate AI into their back-office workflows.

Implications for automation

For businesses, the release of Terra represents a lower barrier to entry for automation. Many firms have hesitated to automate routine administration because of the high costs associated with top-tier models. Terra provides a way to handle bulk tasks like invoice processing more economically.

The model's ability to handle visual reasoning and computer use makes it a candidate for various administrative roles. It can interact with software to move data between different systems, which is a core requirement for many modern offices.

However, because the factuality and computer-use figures come directly from OpenAI, caution is still required. Anyone planning to use Terra for finance or administration should test the model on their own specific documents before relying on it for production tasks.