xAI's flagship Grok 4.6 now runs on all three major clouds: Amazon Bedrock, Microsoft Foundry and Google Cloud Vertex AI. Here is what changed, what each one costs, and when the xAI API is the better route.
1. What is Grok 4.6 on AWS, Azure and Google Cloud?
Grok 4.6, xAI's current flagship model, is now available directly through all three major cloud platforms: Amazon Bedrock, Microsoft Foundry (the renamed Azure AI Foundry) and Google Cloud's Vertex AI, alongside the existing xAI API and the Grok consumer apps.
The three rollouts landed close together. xAI's own news pages, matched by an announcement from each cloud provider, put Amazon Bedrock first on 19 August 2026, Google Cloud's Vertex AI next on 21 August 2026, and Microsoft Foundry last on 26 August 2026, all within about two weeks of Grok 4.6's own launch in August 2026.
None of this changes what Grok 4.6 can do. It is the same model xAI describes as its most intelligent and fastest for chat and code, with the same 500k token context window listed on the xAI API. What changed is where an organisation can call it from, and whose invoice the usage lands on.
2. Is Grok available on Amazon Bedrock?
Yes. Amazon Bedrock added Grok 4.3 in June 2026, then made Grok 4.6 generally available on 19 August 2026, about a week after Grok 4.6 itself launched, according to AWS's own release notes and a matching announcement on xAI's news page.
On Bedrock, Grok 4.6 is listed at $2 per 1 million input tokens and $6 per 1 million output tokens, the same rate as the xAI API, with the same 500k token context window and four configurable reasoning effort settings. Cross-Region inference is available in every AWS region where Bedrock is offered, and Amazon Bedrock in AWS GovCloud (US) added Grok 4.6 support later in August 2026, giving government and other regulated customers a route to the model inside their existing AWS compliance boundary.
Grok 4.6 reaching AWS, Azure and Google Cloud is a distribution change, not a new model, so the real decision is which account should own the bill.
3. Is Grok available on Microsoft Azure?
Yes, in public preview through Microsoft Foundry, the platform formerly called Azure AI Foundry. Microsoft added Grok 4.6 on 26 August 2026, confirmed on both xAI's news page and Microsoft's own Community Hub post, which describes the model as built for long-horizon reasoning and complex agentic workflows.
Deployment goes through the Foundry model catalogue: search for Grok in the model catalog, open its model card, then deploy through standard pay-as-you-go billing or provisioned throughput units for steady production volume. Enterprise teams can authenticate with Microsoft Entra ID and role-based access control instead of managing a long-lived API key, which is the main reason a Foundry deployment appeals to an organisation already built around Azure identity and governance tools.
The one thing worth flagging is price. Third-party trackers disagree on the exact per-token rate and context window Microsoft lists for Grok 4.6 on Foundry: some report the same $2 input and $6 output rate as the xAI API, others describe a shorter context window than the 500k figure quoted elsewhere. Rather than repeat a contested number, check the live Foundry pricing page and the model card inside your own portal before committing a workload.
4. Is Grok available on Google Cloud?
Yes. Grok 4.6 reached Google Cloud through Vertex AI's Model Garden in preview on 21 August 2026, confirmed on xAI's own announcement and Google Cloud's own documentation for partner models.
On Vertex AI, Grok 4.6 pricing starts at $2 per 1 million input tokens and $6 per 1 million output tokens, matching the xAI API. The model keeps its 500k token context window and four reasoning effort settings (low, medium, high and xhigh), and it accepts text and image input with function calling and structured outputs. Access runs through Vertex AI or the newer Gemini Enterprise Agent Platform, so a team already built around Google Cloud IAM does not need a separate xAI account just to start testing.
5. How much does Grok cost through a cloud platform versus the xAI API?
On Amazon Bedrock and Google Cloud Vertex AI, the listed Grok 4.6 rate matches xAI's own API price exactly: $2 per 1 million input tokens and $6 per 1 million output tokens, with the same 500k token context window. That consistency across two independent cloud providers is unusual among third-party model hosts, and it suggests xAI set a matching list price rather than leaving each cloud to set its own margin.
What actually changes is everything wrapped around the token price. Cloud billing can add the provider's own charges on top, such as data transfer, private networking, storage or provisioned throughput, and usage is billed through your AWS, Azure or Google Cloud account rather than a direct xAI invoice. That distinction matters most for procurement: a team required to keep spend inside an existing cloud commitment, or route it through an enterprise agreement already in place, now has a way to use Grok without opening a separate vendor relationship with xAI.
6. Should you use Grok through the cloud or the xAI API directly?
Use the xAI API directly when you want the newest models first and the simplest billing. In every case so far, a cloud rollout has followed the direct API launch by one to three weeks, and xAI's own docs and pricing page remain the fastest place to see a new Grok release land.
Use a cloud platform when procurement, a security review, an existing spend commitment, or an in-house Entra ID or IAM setup make a direct xAI account harder to approve than usage billed through a cloud account you already run. None of the three integrations change what Grok 4.6 can do; they only change which account pays for it and which compliance boundary it sits inside.
7. The bottom line
Grok 4.6 reaching Amazon Bedrock, Microsoft Foundry and Google Cloud Vertex AI within about two weeks in August 2026 is a distribution change more than a pricing one. The model itself, its 500k token context window on two of the three platforms, and its headline price have not moved. What is new is how many procurement routes now lead to the same model.
Takeaway: check which of the three platforms your organisation already has an account and a compliance sign-off for, confirm the live price and context window on that platform's own pricing page before committing a workload, and treat the xAI API's published rate as the reference point whenever a cloud listing looks out of step with it.