OpenAI launches GPT-6.1 Terra with 50% lower API pricing

A model built for action

OpenAI has officially released GPT-6.1 Terra, a new model specifically designed to handle professional workflows that require interacting with computer software. Unlike models that primarily generate text, Terra is built to interpret what is happening on a screen and operate applications to complete tasks. This makes it a primary tool for users who need to move data between systems or perform multi-step digital actions.

The model is being positioned as a faster and more cost-efficient choice for complex work. It can handle tasks such as drafting documents, filling out forms, and interpreting technical drawings. It is also capable of editing video and performing other actions that involve visual reasoning across different software environments.

As reported by EdTech Innovation Hub, the model is now available to a wide range of users. This includes subscribers to ChatGPT Plus, Pro, Enterprise, and Edu, as well as developers who use the OpenAI API. The rollout ensures that both individual professionals and large organizations can integrate these new capabilities into their existing workflows.

Significant cost and speed gains

One of the most notable changes with GPT-6.1 Terra is the pricing structure for developers. The standard API prices are 50% lower than those of the previous model, GPT-5.6 Terra. Specifically, the model costs $1 per million input tokens and $5 per million output tokens, which represents a major shift for those building high-volume applications.

OpenAI also notes that the model is more efficient in how it processes information. It uses fewer tokens to complete certain tasks, which means that users with existing usage allowances can get more work done. This efficiency is backed by benchmarks showing that Terra can complete certain tasks more than three times faster than other high-end models like GPT-6 Astra and GPT-6.1 Sol.

These improvements are particularly relevant for companies trying to scale AI agents. By reducing the cost per task while increasing the speed of completion, OpenAI is making it easier for businesses to deploy AI that performs repetitive computer-based work. The goal is to provide a high-performance option that does not break the budget of large-scale operations.

Benchmarks in reasoning and coding

The performance gains for Terra are clearly visible in several specialized benchmarks. On the Agents' Last Exam, which tests professional computer-based tasks, Terra scored 15 percentage points higher than GPT-5.6 Terra at a medium reasoning level. These tasks include complex actions like converting architectural drawings into 3D models or creating restoration logs from animated footage.

The model also shows significant progress in coding and factual accuracy. On the Terminal-Bench 4.0, which measures how well an AI can solve software engineering tasks using command-line tools, Terra solved 50.3% of tasks at maximum reasoning effort. This is a substantial jump from the 21.5% score recorded by the previous GPT-5.6 Terra model.

OpenAI has also prioritized reducing factual errors during development. In internal tests designed to trick the AI into making mistakes, the factual error rate at low reasoning effort fell from 53.6% to 16.7%. At the highest reasoning levels, the error rate dropped to 4.3%, which OpenAI says is nearly on par with their most powerful systems.

Safety and enterprise readiness

For organizations using ChatGPT Work and Codex, the release is intended to support workflows that require the AI to take actions across multiple applications. This is a shift away from simple text generation toward a system that can interpret visual information and interact with various software tools. It is designed to help with document preparation and software development where information must be pulled from multiple sources.

Safety remains a core component of the new release. OpenAI reported that GPT-6.1 Terra shows improved resistance to prompt-injection attempts, which is when a user tries to bypass the AI's safety rules. The model also shows a better ability to stay within the specific boundaries of the tasks defined by the user.

Under the company's Preparedness Framework, the model meets the Critical threshold for cybersecurity capabilities. It also retains the safeguards that were originally developed for GPT-6 Astra. These measures are intended to give enterprise customers more confidence when they allow the AI to interact with their internal software and data systems.