OpenAI launches GPT-6.1 Terra for affordable computer use

A new balance for AI tasks

OpenAI has debuted GPT-6.1 Terra, a fresh artificial intelligence tool built to handle professional computer-based activities. This model joins the GPT-6.1 family shortly after the release of GPT-6.1 Sol. It is specifically designed to manage digital workflows such as completing browser forms, managing spreadsheets, and editing videos.

The company describes Terra as a model that understands what is on screen and can take actions across different applications. It aims to provide a more capable experience for a user's entire workday by offering the efficiency needed to automate more tasks. This release follows a trend of providing various tiers of models to suit different performance and budget needs.

While it is not intended to replace the most advanced systems, Terra is positioned as a cost-effective alternative. It is meant for users who need reliable results for many professional applications without the high price tag of the company's top-tier models. This helps organizations decide which tasks require high-level reasoning and which can be handled more cheaply.

Speed and visual reasoning

One of the primary advantages of GPT-6.1 Terra is its speed. According to OpenAI, the model can complete computer tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol. This makes it a viable option for businesses that need to automate digital workflows quickly and at scale.

The model also shows strong capabilities in visual reasoning, which is the ability to interpret and reason about visual information. On the MMMU-Pro benchmark, Terra performed nearly as well as both GPT-6 Astra and GPT-6.1 Sol. This suggests it can handle complex visual documents, technical drawings, and other non-textual data effectively.

In software development, the model showed a significant leap over its predecessor, GPT-5.6 Terra. At maximum reasoning effort, it scored 50.3 percent on Terminal-Bench 4.0, which measures an AI's ability to work in terminal-based environments. This is a notable increase from the 21.5 percent score of the previous version.

Drastic reductions in API costs

Pricing is a major component of the GPT-6.1 Terra launch. For those using the developer interface, the model is priced at US$1 for every million input tokens and US$5 for every million output tokens. For developers who reuse content, cached input tokens are even cheaper, costing just US$0.05 per million tokens.

These prices represent a significant discount compared to the flagship GPT-6 Astra. That model is priced at US$10 for million input tokens and US$50 for million output tokens. This means Terra is approximately ten times cheaper for both input and output, which can lead to considerable savings for high-volume applications.

As Tech Edition reported, these lower prices are particularly helpful for companies running AI agents to perform repetitive office tasks. While token prices do not tell the whole story, they provide a lower baseline for interacting with business software. Lowering the cost per request makes it more practical to involve AI in workflows that require many separate actions.

Who can use the new model

OpenAI is making GPT-6.1 Terra available to a variety of users. Eligible paid users can access it through ChatGPT Work and Codex. These platforms are designed for professional activities and software development tasks, respectively, providing a direct path for individuals and organizations to integrate the model.

Developers can also access the model via the API to build their own products and automated systems. This allows for the integration of Terra into custom services that require the model to navigate software or interpret on-screen information. It provides a flexible way to incorporate AI into existing digital infrastructures.

The rollout reflects a broader strategy to make advanced AI capabilities more economical to deploy. By offering a lower-cost alternative, OpenAI allows businesses to allocate resources more effectively. They can use Terra for routine operations like spreadsheet updates while reserving more expensive models for deep research or complex coding.

Performance vs. capability

Despite the impressive benchmarks, OpenAI clarifies that Terra is not the strongest option for every scenario. The company still recommends using GPT-6.1 Sol or GPT-6 Astra for the most demanding coding assignments and complex research tasks. These flagship models remain the preferred choice for workloads requiring the highest level of reasoning.

Benchmark results do not always guarantee identical performance in every real-world situation. Factors such as task complexity, the specific software being used, and the required level of reliability can all influence how the model behaves. Terra is designed to balance performance, speed, and cost rather than maximizing every single metric.

The practical value of the model will depend on whether the cost savings translate into reliable results for specific user applications. For developers building AI agents, Terra offers a route for high-volume operations. It allows for a tiered approach where different models are used based on the sophistication of the decision-making required.