Azure Pushes Claude 4.6 in Microsoft Foundry with 1M-Token Context, 600-Page Inputs, and Flat Pricing

X post and the enterprise positioning

In a March 14, 2026 X post, Azure said Claude Opus 4.6 and Claude Sonnet 4.6 now support a 1M-token context window in Microsoft Foundry, with flat pricing and support for up to 600 images or PDF pages per request. That combination matters because long-context access is often paired with premium surcharges or restricted beta gating. Azure framed the update as a practical simplification for teams building long-running AI workflows rather than as a benchmark-only milestone.

What Microsoft’s own blog adds

Microsoft’s Foundry blog post on Claude Sonnet 4.6 says the model is available in Foundry for coding, agents, and professional work at scale, and explicitly calls out a 1 million token context window (GA) with 128K maximum output. The post also positions Sonnet 4.6 as near-Opus-level capability at a lower cost, with adaptive thinking and effort controls intended to help teams tune quality, latency, and cost tradeoffs.

The same post emphasizes enterprise use cases that benefit from large context: massive codebases, multi-document analysis, long financial models, and extended multi-turn workflows. It also highlights computer-use capabilities, citing a 72.5% score on OSWorld Verified, and argues that Sonnet 4.6 can act across browser-based surfaces, legacy systems, and other environments where direct API integrations do not exist. For organizations trying to automate QA, browser workflows, or document-heavy review processes, that packaging is as important as the raw model size.

Anthropic docs confirm the pricing and input limits

Anthropic’s platform documentation independently confirms the core claims in the Azure post. The context-window guide states that Claude Opus 4.6 and Sonnet 4.6 have a 1M-token context window, and that a single request can include up to 600 images or PDF pages. Anthropic’s pricing page also says Opus 4.6 and Sonnet 4.6 include the full 1M context window at standard pricing, meaning a large long-context request is billed at the same per-token rate as a smaller request rather than triggering a special premium band.

That pricing detail is especially relevant for enterprise adoption. Long-context workflows only become repeatable when teams can predict costs for document review, codebase analysis, or agentic orchestration. If the price curve changes sharply after a threshold, many production use cases stay in pilot mode. The “flat pricing across the full context window” claim reduces that uncertainty.

Why this is a meaningful platform update

This is a high-signal infrastructure change because it turns long-context AI from a specialized experiment into a more straightforward procurement and engineering decision. For teams on Microsoft Foundry, the update combines longer working memory, richer multimodal inputs, and enterprise governance in one stack. That can simplify designs for retrieval-heavy copilots, multi-file analysis agents, and coding workflows that previously had to chunk aggressively or pay special long-context premiums.

Primary sources: X post, Microsoft Foundry blog, Anthropic context docs, Anthropic pricing docs.