Last verified: June 11, 2026. Written and maintained by a team that runs Claude in production on both Microsoft Azure and Google Cloud — this site itself runs on Azure. Updated within days of changes to Foundry model availability or pricing.
Anthropic Claude models are fully available on Microsoft Azure through Microsoft Foundry, including Claude Fable 5 as of June 9, 2026. That sentence is simple; everything after it is not. Foundry Claude bills through a different channel than native Azure models, costs more than the direct Anthropic API for some workloads, is not covered by Microsoft for Startups credits, and has already produced surprise bills large enough to make international tech press. This page is the practical guide: what is available, what it really costs, when Foundry is the right call versus going direct to Anthropic, and how not to get burned.
What is available on Foundry (June 2026)
| Model | Context window on Foundry | Notes |
|---|---|---|
| Claude Fable 5 | 1M tokens | Added June 9, 2026 |
| Claude Opus 4.8 | 200K tokens | Smaller window than direct API |
| Claude Opus 4.7 / 4.6 | 1M tokens | |
| Claude Sonnet 4.6 | 1M tokens | The workhorse tier |
| Claude Sonnet 4.5 / Haiku 4.5 | 200K / 200K | |
| Claude Opus 4.1 | — | Deprecated; retires August 5, 2026 |
Deployment is Global Standard only, in East US2 or Sweden Central. Not supported on Foundry: the Batches API, the Admin and Compliance APIs, standard rate-limit headers, and the Ruby SDK. If your workflow depends on batch processing (50% discount on the direct API), Foundry currently cannot do it — that alone decides the question for many production loads.
What it costs — and the premium tier nobody mentions
Headline pricing matches Anthropic: Opus-tier models at $5 per million input tokens and $25 per million output below 200K input tokens. But Foundry adds a long-context premium the direct API does not have at the same threshold structure: above 200K input tokens, input doubles to $10/M and output rises to $37.50/M. If you run large-context workloads — big document analysis, long agent sessions — Foundry is materially more expensive than going direct. Billing flows through Azure Marketplace, which counts toward Microsoft Azure Consumption Commitments (MACC) — genuinely useful if your company has one, irrelevant if you are small.
The billing trap, plainly
Foundry displays Claude in the same catalog, with the same look, as Microsoft’s native models. It is not billed the same. Claude is third-party marketplace billing, and Microsoft for Startups credits do not cover it. Founders who believed their credits covered Claude have received first invoices of $1,600, $13,000, and more; the issue produced mainstream coverage and a petition to Microsoft leadership in March 2026, and at least one documented case of Microsoft support itself giving the wrong answer. Before you deploy Claude on Foundry:
- Confirm which billing channel a model uses — in the Foundry catalog, third-party models bill via Azure Marketplace even though nothing visually warns you.
- Assume startup credits, sponsorships, and most Azure credit types do NOT apply to marketplace models unless explicitly stated.
- Create a budget with email alerts in Cost Management on day one, scoped to the subscription running Foundry, before the first API call.
- Check your bill within 48 hours of first use. Marketplace charges appear under a separate line item from your Azure services.
Foundry vs the direct Anthropic API: a decision guide
| Your situation | Use | Why |
|---|---|---|
| Enterprise with Azure spend commitment (MACC), procurement requires one vendor | Foundry | Claude spend counts toward the commitment; one bill, Azure IAM |
| Data residency or compliance requires Azure tenancy | Foundry | Runs inside your Azure governance perimeter |
| Startup running on Microsoft for Startups credits | Direct API | Credits do not cover Claude on Foundry — the trap above |
| Heavy batch workloads (content pipelines, bulk processing) | Direct API | Batches API gives 50% off; not available on Foundry |
| Large-context workloads (over 200K input tokens routinely) | Direct API | Foundry premium tier doubles input cost above 200K |
| Solo operator or small business just building | Direct API | Simpler billing, full feature surface, same models |
For current direct-API and subscription pricing on the Anthropic side, see our continuously updated Claude pricing guide — the same maintenance model as this page, on our Google-side sister site.
How we actually run it
This site runs on Azure App Service, and the business behind it runs Claude agents in production daily — which is exactly why this page exists. Our own split: interactive and judgment work on Anthropic subscriptions, bulk pipeline work on the direct API with Sonnet-class models and batching, and Azure used for what Azure is genuinely good at — hosting, networking, and the Microsoft ecosystem itself. When that calculus changes (for example, if Batches lands on Foundry), this page will say so the same week.
Frequently asked questions
Is Claude available on Microsoft Azure?
Yes. Claude models including Fable 5, Opus 4.8, and Sonnet 4.6 are available through Microsoft Foundry as of June 2026, deployed as Global Standard in East US2 or Sweden Central.
Does Claude on Azure cost the same as the Anthropic API?
Below 200K input tokens, yes — same per-token rates, billed through Azure Marketplace. Above 200K input tokens, Foundry charges a premium (input doubles to $10/M; output $37.50/M) that the direct API does not at that threshold.
Do Microsoft for Startups credits cover Claude on Foundry?
No. Claude is a third-party marketplace model and is excluded from startup credits — a documented source of surprise bills. Verify your credit terms before deploying.
Can I use the Claude Batches API on Foundry?
No. The Message Batches API (50% discount on the direct API) is not supported on Foundry as of June 2026, along with the Admin API, Compliance API, and standard rate-limit headers.
When does Foundry make more sense than the direct Anthropic API?
When your organization has an Azure consumption commitment, needs Azure-perimeter governance, or requires consolidated Azure billing. For most small operators, the direct API is simpler and cheaper.
Sources: Anthropic Foundry documentation (platform.claude.com), Microsoft Azure blog announcements for Claude in Foundry, Microsoft Learn Q&A threads on Foundry marketplace billing, InfoWorld and The Register coverage of the Foundry startup-credit billing issue (March 2026). This page is actively maintained; errors are fixed within days.