Grok 4.6 Lands on Microsoft Foundry: xAI's Flagship Goes Enterprise

SpaceXAI has put its flagship model where enterprise buyers actually shop. On August 26, the company announced that Grok 4.6 is now available on Microsoft Foundry, bringing xAI’s frontier model to Microsoft’s managed model catalog through public preview. For an AI lab that spent its first years selling directly to developers, landing a slot in the same enterprise storefront that distributes OpenAI’s models is a quiet but consequential shift in how Grok reaches customers.

The move comes exactly two weeks after Grok 4.6’s debut on August 12, and it targets a specific gap: organizations that want frontier-grade agentic coding and long-horizon reasoning, but need it wrapped in the security, compliance, and governance controls their procurement teams require.

What’s Actually Available

Per xAI’s announcement, Grok 4.6 arrives on Foundry with the full feature set that distinguishes it from the current crop of frontier models:

• 500K token context window — enough to hold an entire codebase, a long research corpus, or a sprawling multi-file refactoring session in a single working memory.
• Configurable reasoning effort — four levels (low, medium, high, xhigh) that let developers trade latency and cost against depth of thought on a per-request basis.
• Text and image input with tool calling and JSON-structured output, per Microsoft’s Foundry Models documentation.
• Chat-completion API surface, deployed as a managed endpoint inside the Foundry model catalog.

The pitch from xAI is straightforward: if you’re building coding agents, engineering copilots, research assistants, or enterprise automation, you can now evaluate Grok 4.6 against other frontier models in a single place, run workload-specific tests, and deploy managed endpoints with enterprise security and governance — without standing up your own inference infrastructure.

One detail worth noting from Microsoft’s documentation: the Foundry deployment is listed in preview with a 200,000-token context window cap, below the 500K maximum available through xAI’s own API. Enterprises planning very-long-context workloads should verify limits for their region and deployment tier before committing.

Why Grok 4.6 Earned the Slot

Grok 4.6 was built with a clear thesis: long-running agents and ambitious interactive work. According to xAI’s launch materials, the model “stays with complex tasks across many steps, whether researching a topic, analyzing information, working across a codebase, or turning an idea into a polished application or work artifact.”

The benchmark story backs that up:

Benchmark	Grok 4.6 High	GPT-5.6 Sol Max	Fable 5 Max
AA Intelligence Index	61	61	62
GDPVal-AA v2	1753	1728	1741
CursorBench v3.2	69.9%	67.2%	70.5%
FrontierCode v1.1	61.3%	60.6%	63.6%
APEX-Agents	57.5%	56.7%	59.2%
Terminal-Bench v3.0	26%	34.6%	34.1%

It matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (a composite of nine benchmarks) and posts the best score on GDPVal-AA v2, a knowledge-work benchmark. Anthropic’s Fable 5 Max holds narrow leads on agentic coding suites, while GPT-5.6 Sol dominates Terminal-Bench. The picture is a genuinely contested frontier — which is exactly why Foundry’s “evaluate side by side” positioning matters.

Training-wise, xAI says 4.6 underwent a longer supplemental run than 4.5, using curated model-generated data for reasoning, high-quality engineering data, an improved optimizer, and regenerated SFT trajectories built by Grok 4.5 itself across reasoning efforts and domains. The RL stage spanned kernel optimization, web development, computer-aided design, and other domain-specific environments.

Pricing through xAI’s own API starts at $2 per million input tokens and $6 per million output tokens, with a fast variant at twice the price. Foundry pricing follows Azure’s standard metered model.

The Strategic Read

Three things make this more than a routine catalog listing.

First, distribution is the current battleground. Model quality at the frontier has compressed — Grok 4.6, GPT-5.6, and Fable 5 all sit within a point or two of each other on composite indices. When capability gaps shrink, reach decides winners. Microsoft Foundry is one of the two or three storefronts through which large enterprises actually consume frontier AI. Being there means Grok is now a line item procurement can approve without exception requests.

Second, it extends the Microsoft–xAI relationship beyond hosting. Microsoft already booked billions in paper gains on its Anthropic stake and operates one of the largest AI infrastructures on the planet. Adding Grok 4.6 to Foundry Models — the same catalog that carries OpenAI’s GPT family — formalizes xAI as a first-party-adjacent supplier rather than a curiosity. Notably, the model is even listed under “Foundry Models sold by Azure,” the billing tier where Microsoft resells models directly.

Third, it’s timed with the agent wave. The same week, xAI expanded Grok Bot — its always-on cloud-computer agents — to Cursor Pro and all Cursor Teams plans. Enterprise buyers reading both announcements see a coherent story: frontier reasoning model (Grok 4.6) plus persistent agent runtime (Grok Bot), now with an enterprise consumption path for the former and a team plan path for the latter.

Caveats and Open Questions

The preview label is doing real work. Context window caps, rate limits, and regional availability all remain deployment-specific per Microsoft’s docs, and “public preview” in Azure parlance means no SLA guarantees yet. Buyers with hard compliance requirements will wait for GA.

There’s also the question of whether the 200K cap in Foundry diverges from the headline 500K elsewhere — a discrepancy that will either close by GA or become a persistent differentiator pushing heavy context users to xAI’s direct API.

And the competitive context is unforgiving. Anthropic’s Fable 5 family leads several agentic benchmarks, and OpenAI’s models have the deepest Microsoft integration by far. Grok 4.6 on Foundry gets xAI into the room; it doesn’t win the room.

What It Means

For enterprise architects, the practical takeaway is simple: the frontier-model menu inside Microsoft Foundry just got one entry longer, and that entry is competitive on agentic coding and knowledge work at an aggressive price point. Side-by-side evaluation — Foundry’s core workflow — is now the cheapest it’s ever been to test whether Grok 4.6’s long-horizon agent stamina holds up on your workloads.

For the industry, it’s another sign that 2026’s frontier race is less about who has the single best model and more about who shows up in every channel: consumer apps, developer tools, direct API, and now the enterprise catalog. Grok 4.6 just checked the last box.