Microsoft Foundry is adding request-time token enforcement, project-level cost attribution, and ROI tracking to help organizations control AI agent spending. The update extends Microsoft’s effort to turn AI agent spending into a FinOps problem by moving governance closer to the requests that generate consumption.

Source

Microsoft says its own supply-chain transformation has reduced cycle times by up to 75% in selected workflows after deploying more than 100 purpose-built AI agents. The result strengthens Azure’s pitch as an end-to-end platform for enterprise AI, combining models, data, infrastructure, applications, developer tools, security, and governance rather than treating each project as a standalone model deployment.

Source

Microsoft is positioning Azure Container Apps Sandboxes as the execution layer for production AI agents, while Microsoft Foundry handles identity, governance, tracing, and evaluation. The separation addresses a growing enterprise problem: agents that can reason and use tools need more than a model and a host application—they need isolated runtimes that can act without gaining uncontrolled access.

Source

Anthropic's Claude Opus 5.5 is now generally available in Microsoft Foundry, hosted on Azure infrastructure. Customers can choose Global Standard or US DataZone deployments. Pricing matches Anthropic's own list, following the release that made Claude Opus 5.5 cheaper than Opus 5 on input, output and cache reads. Anthropic is also shipping new beta API features for long-running agents alongside the model.

Source

Microsoft has made GPT-6 Sol and GPT-6 Luna generally available in Microsoft Foundry. Both models join GPT-6 Astra with Standard deployments across all 28 Global regions and in the US and EU Data Zones, and they come with Foundry-specific pricing that charges more for data residency. The rollout follows OpenAI's launch of the two cheaper tiers, when GPT-6 Sol and Luna cut API costs by about half compared with their GPT-5.6 predecessors.

Source