Grok 4.7 reaches AWS GovCloud, a month after its commercial Bedrock debut

Amazon says Grok 4.7 is now available on Amazon Bedrock in AWS GovCloud (US). The model, which the announcement credits to SpaceXAI, is aimed at coding, agentic tasks and knowledge work. It reaches the isolated US government cloud about three weeks after its commercial Bedrock debut.

What the GovCloud announcement actually says

The GovCloud notice is short. According to Amazon, Grok 4.7 improves on Grok 4.6 in three areas: handling documents that mix content types, repo-scale coding with planning and error recovery, and browser-use agents that fill in forms and navigate portals. Customers reach it through the Bedrock console or the supported Bedrock APIs.

The page does not state a context window, token prices or the exact endpoints and inference profiles available in GovCloud. It sends readers to the Bedrock documentation for supported regions, pricing and the model card instead. That matters because the commercial and GovCloud offerings are separate deployments, and nothing in the notice confirms that every commercial feature carries over.

The commercial launch gives a fuller spec sheet

AWS published a launch blog post on September 28 for the commercial regions, a week after xAI released the model on September 21. That post describes a 500K-token context window, text and image input with text output, and tool calling. Reasoning is always on, with four effort levels: low, medium, high and xhigh, the last being the highest setting. The default is high.

It also lists four APIs (Responses, Chat Completions, Converse and InvokeModel) and two cross-Region inference profiles, us.xai.grok-4.7 and global.xai.grok-4.7. Pricing is pay-per-token with default, priority and flex service tiers. Whether the GovCloud version matches any of this is not established by the GovCloud notice. Agencies should check the model card before assuming it does.

Better scores, but roughly twice the output tokens

The launch post includes Artificial Analysis figures, measured at xhigh effort. Grok 4.7 scores 46 on the Intelligence Index against 44 for Grok 4.6, and 56 on the Coding Agent Index against 47. Its hallucination rate on the AA-Omniscience test is 29%, down from 34%.

The post itself notes the extra tokens, so the model works longer per task. On a pay-per-token service, that can erase part of the benefit if the per-token price is unchanged. These are figures from Amazon’s launch post, not from our own tests, and xAI’s own benchmark claims (including CursorBench and Terminal-Bench) come without numbers in that post.

Who can use it now, and what changes for them

For US public-sector teams and contractors that run workloads in GovCloud, the practical change is a second major frontier model family option inside the same boundary, accessed through APIs they already use for other Bedrock models. Teams that were blocked from using Grok because their data must stay in GovCloud now have a route to try it. Those that want a spec sheet should expect to rely on the commercial launch details until the GovCloud model card says otherwise.

For developers building agents, the improvements Amazon highlights are the ones that matter in government back-office work: filling in forms, working through portals and reading mixed documents. Amazon does not name a customer, a compliance level or an authorization status for the model in this notice, so security teams will need to verify those points themselves. The announcement also does not say whether the faster priority or cheaper flex tiers exist in GovCloud.

The release adds to a run of xAI products reaching enterprise channels. Last week we reported that Grok Voice 3.0 topped Artificial Analysis’ speech-to-speech index, though that is a separate product line from the text model covered here.

Still missing from the official notice: prices, the context window in GovCloud, regional endpoints and any independent evaluation. Until the documentation links fill those in, the announcement confirms availability but not terms.