GPT-6 Astra reaches general availability in Microsoft Foundry

Microsoft published the availability post on 3 September 2026, describing GPT-6 Astra as OpenAI’s newest frontier model and saying it is generally available for all customers in Microsoft Foundry. The framing is agentic rather than conversational: the post says the model can take an open-ended challenge, reason through it in multiple steps, create a plan, and produce a polished result.

The numbers

Pricing is published per million tokens, and it splits by context length before it splits by geography. On Standard Global, short context input is $10.00, cached input $1.00, cached writes $12.50 and output $50.00. Long context on the same deployment runs $20.00 input, $2.00 cached input, $25.00 cached writes and $75.00 output, so moving to long context doubles the input rate and adds half again to output.

The US Data Zone carries a flat surcharge against Global at every point in that table: $11.00 and $55.00 for short context, $22.00 and $82.50 for long. Microsoft applies the same margin to dedicated capacity, writing that US Data Zone Provisioned Throughput is priced at a 10% premium to Global Provisioned Throughput. Standard is described as pay-as-you-go for variable demand, Provisioned Throughput as dedicated model-processing capacity for workloads that need consistent latency and guaranteed throughput.

What the post claims, and what it hedges

Microsoft says computer use lets the model interpret on-screen information and interact with approved interfaces, including workflows without dedicated APIs, and it names software engineering, Power BI dashboards, document production and customer record updates as scenarios. The benchmark claim is hedged in the post itself: OpenAI reports state-of-the-art results on selected computer-use evaluations, and Microsoft adds that performance varies by task, tools, configuration, and safeguards. The post is blunter about the risk of an agent reading a screen, noting that displayed content may be incomplete, misleading, or designed to influence behaviour, and pointing customers at scoped credentials, approved resources and human checkpoints. On data handling it is direct: prompts and outputs are not used to train the models.

Where this sits against Copilot

This is a Foundry announcement, so no Copilot seat changes today. Our ledger prices Copilot by the seat: Microsoft 365 Personal at $9.99 a month, Family at $12.99, Premium at $19.99, with image generation capped at 60 credits per month on Personal and Family and agent work capped at 25 tasks per month shared across agents. Premium buys higher usage limits rather than a published rate per request. That is the choice in front of a buying team. Take Astra directly at the Foundry meter above and carry the identity, approval and monitoring work yourself, or wait on a seat price and accept whatever Microsoft puts behind the Copilot box.