OpenAI launches GPT-6.1 Terra to lower costs for AI agents

A new model for visual tasks

OpenAI has officially released GPT-6.1 Terra, a reasoning model specifically engineered for high-volume visual and computer-use work. This new model replaces the previous GPT-5.6 Terra and is now live in the API under the name gpt-6.1-terra. It is designed to handle complex inputs like text and images while returning text outputs.

The model features a context window of 1,050,000 tokens, which allows it to process large amounts of information in a single session. Users can also generate up to 128,000 output tokens. Its knowledge cutoff is set to June 1, 2026, ensuring it has access to relatively recent information.

One of the primary goals for Terra is to provide a balance between speed and intelligence. It is built to perform tasks such as matching product photos to catalog entries or interpreting technical drawings. It is available as a closed-weights model, meaning it can only be accessed through OpenAI-hosted platforms like the API, ChatGPT Work, and Codex.

How the reasoning works

GPT-6.1 Terra utilizes a variable reasoning effort system that allows users to choose between none, medium, and max settings. Medium is the default setting for most interactions. This flexibility allows developers to scale the model's thinking power based on the complexity of the task at hand.

The model supports a variety of tools through the Responses API to interact with the digital world. These include computer use, a hosted shell, code interpreter, web search, file search, and MCP. These capabilities enable the model to perform actions on a computer rather than just generating static text.

When performing these tasks, Terra is noted for its speed. According to the launch details, OpenAI says Terra completes tasks over 3 times faster than GPT-6 Astra and GPT-6.1 Sol. This efficiency is a key differentiator for developers who need to run many automated processes in a short period of time.

Performance and benchmarks

OpenAI evaluated Terra on several benchmarks to see how it handles professional computer tasks. On the Agents’ Last Exam v1, which involves turning architectural drawings into 3D models, Terra scored 57.6% at medium effort. This matches the score of the more expensive GPT-6 Astra, but at a much lower cost of $0.29 per task compared to $4.82.

In visual reasoning tests like MMMU-Pro, Terra reached a score of 85.5% at max effort. This puts it within 1.2 points of the GPT-6 Astra model. While it is not the absolute fastest at coding, it showed significant improvements over its predecessor, with the error rate on internal adversarial prompts dropping from 53.6% to 16.7% at low effort.

However, there are still some limitations in specific areas like coding. On the Terminal-Bench 4.0, Terra solved 50.3% of tasks at max effort, which is behind Claude Sonnet 5.5. Because of this, OpenAI recommends using the Sol or Astra models for the most complex coding requirements while using Terra for high-volume visual work.

Cost math for developers

The pricing structure for GPT-6.1 Terra is a significant shift for teams automating screens, forms, and documents at scale. It costs $1 per million input tokens and $5 per million output tokens. This is exactly half the cost of the GPT-6.1 Sol model, which costs $2 and $10 respectively.

For even larger scale operations, OpenAI offers Batch and Flex options that are 50% cheaper than the standard pricing. There is also a fast mode available, though it costs twice the standard rate. For very long prompts exceeding 272K input tokens, the billing rate increases to 2x for input and 1.5x for output.

When compared to competitors, Terra offers a middle ground in the market. It is significantly cheaper than the GPT-6 Astra, which costs one-tenth as much per million tokens. It also sits at a lower price point than Claude Opus 5.5, which is listed at $4 per million input and $20 per million output. This positioning makes it a primary choice for high-volume agentic workflows.

Availability and next steps

Users can begin using GPT-6.1 Terra immediately through the OpenAI API. It is also integrated into the ChatGPT interface for users with Plus, Pro, Enterprise, and Edu subscriptions. These users can access the model specifically within the ChatGPT Work and Codex environments.

The release marks a move toward making