1300 166 138 Tell us what you need
News / xAI

Amazon Bedrock adds Grok 4.7 as a rival to Claude and GPT

By Monika Tantau • Published 2 months ago
Amazon Bedrock adds Grok 4.7 as a rival to Claude and GPT

Key takeaways

• AWS added xAI's Grok 4.7 to Amazon Bedrock on 28 September, the same day as Claude Sonnet 5.5
• Global pricing is $2.85 per million input tokens and $8.55 output, the cheapest output of the new arrivals
• 500K token context, text and image input, four reasoning levels and OpenAI compatible APIs
• No Australian profile: the global option routes requests worldwide, unlike Claude Opus 5.5

Most Read

Amazon Web Services added Grok 4.7 to Amazon Bedrock on 28 September, according to the AWS What’s New page. xAI’s newest frontier model joins the platform with a 500K token context window, text and image input and four reasoning levels, and it arrived on the same day as Claude Sonnet 5.5.

That makes it a busy week for Bedrock. Claude Opus 5.5 and GPT 6 Sol landed on 22 September, Sonnet 5.5 and Grok 4.7 on 28 September. An Australian team with an AWS account can now pick between the latest models from Anthropic, OpenAI and xAI without signing a single new vendor contract.

What Grok 4.7 brings to Bedrock

AWS describes Grok 4.7 as a model built for coding, agentic tasks and knowledge work. Compared with Grok 4.6 it handles mixed document sets better, is more dependable on repo scale coding with planning and error recovery, and runs browser use agents that fill forms and navigate portals.

The AWS launch post quotes Artificial Analysis figures: an Intelligence Index of 46 against 44 for Grok 4.6, a Coding Agent Index of 56 against 47, and a hallucination rate of 29 per cent against 34 per cent. It also uses roughly twice as many output tokens per task, about 81,000 against 38,000, which is xAI’s way of saying it keeps working on hard problems rather than answering fast.

What it costs next to Claude and GPT 6

On Bedrock’s global inference profile, Grok 4.7 costs $2.85 per million input tokens and $8.55 per million output tokens, with cached input at $0.71. The US only profile is 10 per cent dearer at $3.13 and $9.40. Priority service is billed at 1.75 times those rates and Flex at half. All figures are converted from US dollars at the Reserve Bank’s 28 September rate.

That output price is the lowest of the four new arrivals. GPT 6 Sol on the same global profile is $2.85 in and $14.25 out. Anthropic lists Sonnet 5.5 at $2.81 and $14.04, and Opus 5.5 at $5.62 and $28.08. For agent workloads that generate a lot of text, Grok 4.7 undercuts Sonnet 5.5 on output by about 40 per cent, at least on paper. Our Claude Sonnet 5.5 coverage has the detail on what the Anthropic model changed.

The data residency catch

The data residency catch

Here is the part Australian buyers need to read twice. Grok 4.7 has no in region option and no Australian geo profile. AWS offers it two ways only: a US profile that keeps requests inside four American regions, and a global profile that, in AWS’s own words, routes your request to a supported commercial region worldwide, for use when you have no data residency constraints.

You can call the global profile from Sydney or Melbourne, so the model shows up in the console here. But the tokens may be processed in Virginia, Frankfurt, Mumbai or São Paulo. Claude Opus 5.5, by contrast, has an Australian geo profile that AWS says keeps data within Australian regions. Sonnet 5.5 and GPT 6 Sol are global or US only, like Grok. Our data residency guide covers what that means under the Privacy Act and the US CLOUD Act.

Why Australian businesses should care

For a Melbourne software team building internal coding agents on public repositories, Grok 4.7 is the cheapest frontier option on Bedrock today, and it speaks the OpenAI compatible Responses and Chat Completions APIs, so swapping it in is a config change. The 500K context also covers most codebases without chunking.

For anyone handling customer records, health data or anything an APRA regulated entity touches, the global profile is an offshore transfer by design. That does not rule Grok out, but it moves it from a quick trial to a privacy impact assessment. Where we do AI strategy work with regulated clients, the model shortlist starts with where the tokens are processed, not with benchmark scores.

Four vendors, one bill

In seven days Bedrock added the flagship models of Anthropic, OpenAI and xAI. Model choice is turning into a line in a config file, billed through the AWS account a business already has, and the switching cost is heading toward zero.

What has not converged is where the data goes. Until xAI or AWS adds an Australian profile, Grok 4.7 is a strong option for code and public content, and a question mark for everything else. Expect that gap, not raw capability, to decide which of the four Australian enterprises standardise on.

Scroll to Top