Grok 4.7 is on Amazon Bedrock: 500K context, US Geo and Global inference
AWS What’s New, 28 September: SpaceXAI Grok 4.7 for coding, agents and knowledge work. ML blog: four reasoning levels, default high; us.xai.grok-4.7 and global.xai.grok-4.7. Builds on 4.6’s mixed documents, repo-scale coding and browser-use agents.
Feature2 minIn-depth
Priya Shah2 min read
- Grok 4.7
- Amazon Bedrock
- AWS
- SpaceXAI
- Agents

Photo: Amazon Web ServicesSource page
Amazon Bedrock now serves SpaceXAI’s Grok 4.7. AWS posted the What’s New note on 28 September. The September 21 launch of the weights — long-horizon coding, a new safeguard stack — is already on this site. This file is the cloud SKU: US Geo and Global cross-Region inference, wired into Bedrock’s catalog for coding, agentic work and knowledge tasks.
What’s New says 4.7 builds on 4.6 with better mixed-document handling, more dependable repo-scale coding with planning and error recovery, and stronger browser-use agents for form fills and portal navigation. The machine-learning blog, by Suheel Farooq, Anirban Gupta, Fabio Branco, Ikenna Izugbokwe and William Yap, puts numbers on the package: a 500,000-token context window, and reasoning effort at low, medium, high and xhigh. Default is high. xAI’s line, via AWS, is that this is its most capable model for coding and knowledge work — it works longer on hard tasks and checks its own output more carefully before moving on. It was also trained to understand the Grok Bot harness natively.
On Bedrock the model is reached through inference profiles, not a bare ID: us.xai.grok-4.7 keeps processing in the US geography; global.xai.grok-4.7 can land in any supported commercial Region, cheaper, with less control over where a given request is served. AWS lists the Responses, Chat Completions and Converse APIs. Text and image in, text out. Guardrails, implicit prompt caching and structured outputs attach the usual Bedrock way. Service tiers are default, priority and flex. AWS says to set effort on purpose: leaving the default on a short classification call will spend reasoning tokens the job does not need.
Artificial Analysis numbers in the same blog are a second reference, not xAI’s own card. Intelligence Index 46 against 4.6’s 44; Coding Agent Index 56 against 47; AA-Briefcase Elo 1,657 against 1,546. The trade-off AWS printed: about 81,000 output tokens per Intelligence Index task, against about 38,000 for 4.6, measured at xhigh for 4.7. That is why the effort knob is in the launch post. The Bedrock note does not replace the September 21 safety write-up. It is where the model sits if you already buy inference from AWS.


