Microsoft Foundry model retirements in November 2026: Claude 4.5 and the o-series

Microsoft's lifecycle policy defines a retired model as removed from service: all inference requests return 410 Gone, existing deployments stop working and no new deployments can be created. Retirement dates can't be extended.

For Global Standard, Data Zone Standard and Standard deployments, Microsoft upgrades the deployment to the listed replacement at retirement where an upgrade is supported, on a rolling, region-by-region schedule. A deployment whose versionUpgradeOption is NoAutoUpgrade stops working at retirement. Provisioned deployments are never upgraded automatically: you create a deployment of the new model and move your traffic before the date.

Where Microsoft lists no replacement, as for claude-haiku-4-5 and claude-opus-4-5, there is no upgrade target, so calls to those deployments fail with 410 Gone from the retirement date until your code points at another model.

The o-series models are already in the Deprecated stage: subscriptions that have deployed that model version can still create and manage deployments until 19 November, and other subscriptions can't deploy them at all.

Microsoft's guidance is to evaluate candidate models against your own prompts and representative data, comparing quality, latency and cost together, and to confirm API compatibility, feature support, deployment-type and regional availability, capacity and quota before you switch. The listed replacement is Microsoft's recommendation and does not limit your choice.

Microsoft manages Claude quota at the subscription level per model and version. All Global Standard deployments of one model share a single pool across regions, and Data Zone Standard deployments share a pool per data zone; Microsoft lists Claude Data Zone Standard for the US data zone only. A move from a 4.5 model to a 5.5 model therefore draws on a different pool, and on pay-as-you-go the default for Sonnet and Haiku 5.5 is half the 4.5 default:

The Quota page in the Foundry portal shows what your subscription has. To go beyond the defaults, submit Microsoft's quota increase request form. Microsoft evaluates each request individually and does not guarantee approval, so file it well before the retirement date.

For Azure OpenAI Standard deployments that Microsoft upgrades automatically, quota is carried over for you. For Provisioned deployments you must make sure quota for the target model is available before you migrate; provisioned throughput units are interchangeable across provisioned deployments.

Standard Compute is one option. It puts Claude, GPT, Gemini and leading open models behind one API, and Claude is served across several clouds, so one provider's quota or outage doesn't stop your work. If a provider goes down, your request fails over to another approved provider of the same model automatically, without leaving your region or provider rules.

Set a Europe rule for your company, a team or a single key, and every request, including retries and fallbacks, is served only by approved European inference endpoints. If no European endpoint can serve it, the request is refused before your prompt is sent. Every person and app gets its own key with a fixed monthly budget that never overruns.