Microsoft (NASDAQ: MSFT) has made OpenAI GPT-6.1 Terra generally available in Microsoft Foundry, putting a visual-reasoning and computer-use model into the company’s enterprise AI platform. The move gives organizations a production option for agents that need to inspect documents or screens and then act inside software, while keeping identity, networking, governance and deployment concerns within Azure.
Microsoft describes GPT-6.1 Terra as a model for day-to-day professional tasks with stronger visual reasoning, faster computer use and natural-language writing. It accepts text and images, supports up to one million input tokens, and lets developers tune reasoning effort and response verbosity. The model can also send interim updates during a long-running task when the orchestration layer supports that behavior, according to Microsoft.
Those capabilities target a common enterprise gap between recognizing information and completing a workflow. A model may identify details in a product photo, technical drawing, floor plan or application screen, but the business value comes only when that finding can drive a controlled next action. Microsoft cites examples such as filling forms, drafting proposals, reviewing technical visuals, updating catalog records and navigating software workflows.
The general-availability status matters because teams can now place that kind of model work inside their existing Foundry development path instead of treating it as a separate experiment. Microsoft said GPT-6.1 Terra is available through the Responses API without new SDK requirements or schema changes. That lets developers use the same platform interfaces for a model whose primary role is high-volume visual and screen-based work, rather than forcing a different integration pattern for each specialized capability.
Deployment geography is part of the enterprise case. Microsoft said the model is available in Global Standard deployments across supported regions and in Data Zone Standard deployments in the United States, European Union and Asia-Pacific. The company positions those choices as a way for customers to adopt the model in the environment where they already build and scale applications without giving up data-residency options. Organizations still need to verify their own selected regions, quotas and data-handling obligations before deployment.
Microsoft also differentiates GPT-6.1 Terra from GPT-6 Astra. In the company’s description, Astra is suited to open-ended, multistep reasoning and planning, while Terra handles high-volume visual and screen-oriented tasks at a lower price point. That division is relevant to enterprise architecture: a system can reserve broader planning work for one model and direct repeated visual tasks to another, rather than treating every request as a single premium inference job. It can also help teams align cost and latency decisions with the actual task instead of defaulting every workload to a single model.
The addition does not turn computer use into an unattended business process. Teams must decide which actions require approval, evaluate reliability on their own data and maintain audit trails around access and changes. Still, Microsoft Foundry’s addition of GPT-6.1 Terra signals a more practical model-selection pattern than earlier one-model AI stacks. Enterprises can increasingly separate planning, visual interpretation and action-oriented work while keeping model access, evaluation and governance in one platform.