OpenAI launches GPT-6.1 Terra for visual reasoning and computer use

A new model for visual logic

OpenAI has introduced GPT-6.1 Terra, a model built to handle visual reasoning and precise computer use. It is designed to be fast and efficient while maintaining the ability to produce clear, natural writing. This model aims to bridge the gap between simple text generation and complex visual understanding.

The model can perform several specific types of visual tasks. It is capable of matching product photos to catalog entries and interpreting complex technical drawings. It can also act on application screens, making it a candidate for automated software navigation.

Users can also use the model to draft and revise text content. Because it supports visual reasoning, it can process images as input while providing text as the primary output. This makes it versatile for workflows that require a mix of sight and language.

Context and reasoning controls

One of the primary technical specifications for GPT-6.1 Terra is its 1,050,000 context window. This refers to the amount of information the model can keep in its immediate memory during a single interaction. A large window allows for the processing of very long documents or many images at once.

The model also introduces a specific parameter called reasoning_effort. This allows users to choose how much computational power the model spends on thinking. The available settings include none, low, medium, high, xhigh, and max, with medium being the default choice.

In addition to the large context window, the model supports a maximum output of 128,000 tokens. This ensures that while the model can read a vast amount of data, it can also generate substantial responses in a single turn. These features are intended to help users manage complex, multi-step projects.

Tool integration and capabilities

GPT-6.1 Terra is equipped with a wide array of tools when used with the Responses API. These capabilities include web search, file search, and image generation. It also supports a hosted shell and a code interpreter, which allows the model to execute code to solve problems.

The model is specifically noted for its computer use capabilities. This means it can interact with software interfaces to perform actions on a user's behalf. It also supports MCP, which is a standard for connecting models to various data sources and tools.

For developers, the model supports structured outputs and function calling. It can also be used for tasks like translation, transcription, and speech generation. However, it does not support audio input or video input at this time, focusing instead on text and image modalities.

Pricing and usage tiers

OpenAI has established a detailed pricing structure for GPT-6.1 Terra based on token usage. Input text tokens cost $1.00 per 1 million tokens, while output tokens cost $5.00 per 1 million. Cached input tokens are significantly cheaper, priced at 5% of the uncached input rate.

There are specific costs associated with different types of retention and speed. Cache writes for 30-minute retention cost $1.25 per million tokens, while 12-hour retention costs $2.50. Users can also opt for Batch or Flex pricing, which are 50% lower than standard rates.

The model also includes premiums for certain features. Regional processing can add a 10% premium where available, and Fast mode prices are double the standard rate. Additionally, prompts exceeding 272K input tokens incur a 2x multiplier on input and cache rates, and a 1.5x multiplier on output rates.

Technical limits and deployment

To ensure reliable access, OpenAI uses rate limits to cap requests, tokens, and audio duration. These limits are tied to the user's usage tier, which automatically increases as a user spends more on the API. This system helps maintain performance for all users during high-demand periods.

The model is available through several endpoints, including Chat Completions and the Responses API. It also supports the Assistants API and fine-tuning, though fine-tuning is listed as not supported for this specific model. Developers can find migration guidance in the GPT-6 guide.

The knowledge cutoff for GPT-6.1 Terra is June 01, 2026. This date marks the limit of the information the model was trained on. For real-time information, users must rely on the supported web search tools provided within the model's ecosystem.