Microsoft Brings OpenAI’s GPT-6 Astra to Foundry With Limited Access

Microsoft has begun rolling out GPT-6 Astra, OpenAI’s newest frontier model, through the Microsoft Foundry Limited Access Program, with availability expanding to participating customers over the coming days and consumption-based pricing starting at $10 per million input tokens under Standard Global deployment, according to the Microsoft Azure Blog announcement.

The announcement, authored by Steve Sweetman and Naomi Moneypenny, describes Astra as a model designed to take an open-ended challenge, reason through it in multiple steps, create a plan, and produce a finished result. The post frames the model around three workplace functions: deliberate planning and decision support, in which the model breaks a challenge into steps, evaluates options, communicates a recommendation, and identifies next actions for review; polished output, in which it applies context, templates, and quality standards across a workflow to produce documents, spreadsheets, presentations, and analyses ready for review; and execution across applications, using tool use and computer use to complete multi-step tasks with human oversight.

Computer Use and Enterprise Controls

Astra’s computer-use capabilities are designed to work across familiar applications, including workflows without dedicated APIs. According to the Azure post, the model can interpret on-screen information and interact with approved interfaces to support tasks such as updating records, navigating development tools, testing software, and assembling results into reports. The post notes that OpenAI reports state-of-the-art results on selected computer-use evaluations, while cautioning that performance varies by task, tools, configuration, and safeguards.

Microsoft pairs those capabilities with containment controls. Because content displayed in an application may be incomplete, misleading, or designed to influence an agent’s behavior, the post says Foundry helps customers define access, approvals, and monitoring, and design workflows with scoped credentials, approved resources, human checkpoints for consequential actions, and activity records aligned to their risk requirements.

The Azure post lists enterprise scenarios Microsoft says it is seeing, including software engineering work in which Astra reproduces complex bugs, investigates likely causes, proposes fixes, and prepares changes for developer testing and review; business intelligence work in which it builds and refines dashboards in Power BI; professional document production following existing templates and business standards; and application workflows such as updating customer records, processing forms, and testing websites where dedicated APIs are limited.

On governance, the post states that Foundry complements OpenAI’s model-level work with Microsoft Entra identity and access management, encryption in transit and at rest, private networking options, role-based access controls, content filtering, safety evaluations, monitoring, and governance tools. Prompts and outputs are not used to train the models. The post adds that these capabilities help customers configure safeguards and maintain oversight but do not eliminate risk or replace each organization’s responsibility to select and configure controls appropriate to its scenarios and regulatory obligations.

What OpenAI Reported at Launch

OpenAI’s own launch post for GPT-6 Astra says the model is rolling out to a limited set of organizations and will become available over the coming days to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock. OpenAI describes Astra as its most aligned model and reports state-of-the-art results across computer use, browsing, software engineering, cybersecurity, science, and professional work evaluations.

Among the figures OpenAI reported: Astra scored 57.9% on Terminal-Bench 4.0, a coding evaluation, compared with 37.3% for GPT-5.6 Sol; 72.6% on an offline OSWorld 2.0 computer-use set, compared with 65.7% for GPT-5.6 Sol; and 96.0% on GPQA Diamond, a graduate-level science benchmark. OpenAI also reported that in latency simulations on OSWorld 2.0, Astra achieved higher computer-use performance in about 47% less time per task than GPT-5.6 Sol, and that combined with an updated Codex harness this translates to 1.9x faster task completion on the Mind2Web benchmark. These are OpenAI’s self-reported evaluation results.

On safety, OpenAI said in a September 1, 2026 update that it assesses Astra as meeting the Critical cybersecurity capability threshold under its Preparedness Framework — the first model the company has designated at that level — meaning that with the right tools and access it can find previously unknown security flaws and develop ways to exploit them across well-protected systems without step-by-step human guidance. OpenAI said it delayed parts of Astra’s development and release over several weeks while strengthening protections, and that the version launching now refuses more advanced cybersecurity tasks such as creating proof-of-concept exploits, with expanded defensive access planned through its Daybreak program in the coming weeks. The Azure post similarly notes that OpenAI describes Astra as its most aligned model to date and plans to publish supporting alignment, safety, and computer-use evaluations in its launch materials.

Deployment and Pricing

GPT-6 Astra will be available through Standard deployments including Global and U.S. Data Zone options, and customers can choose the deployment based on workload requirements. Microsoft describes the consumption-based model as a way for teams to begin building without committing to reserved capacity, and says Astra is designed for token efficiency on complex work, while noting that actual usage and costs will vary by workload and configuration.

The published pricing table lists Standard Global short-context deployment at $10.00 per million input tokens, $1.00 per million cached input tokens, $12.50 per million cached writes, and $50.00 per million output tokens. Standard Global long-context deployment is listed at $20.00 input, $2.00 cached input, $25.00 cached writes, and $75.00 output. Standard Data Zone (US) pricing is $11.00 input, $1.10 cached input, $13.75 cached writes, and $55.00 output for short context, and $22.00 input, $2.20 cached input, $27.50 cached writes, and $82.50 output for long context.

Customers can evaluate the model through the Foundry Models catalog and build cross-application workflows with the Foundry Agent Service, according to the Azure post. The post also includes statements from launch partners: Replit CTO Luis Hector Chavez said GPT-6 Astra through Microsoft Foundry adds agentic capability beyond code generation, and Albertsons Companies VP of Data and AI Anirban Nandi said Azure OpenAI on Microsoft Foundry helps the grocer balance speed of adoption with security, governance, and operational controls.