Grok 4.6 is generally available on Amazon Bedrock

xAI’s latest flagship is generally available on Amazon Bedrock. On August 19, 2026, both xAI and AWS said Grok 4.6 is live for Bedrock customers—an enterprise access story, not a new-model announcement.

The model was already xAI’s flagship for long-running agents. What changed is that teams can now call it through Bedrock, with US and global cross-Region inference, a 500K-token context window, configurable reasoning up through xhigh, and listed prices of $2 per million input tokens and $6 per million output tokens.

Built for long agent sessions

xAI calls Grok 4.6 its latest flagship, built for long-running agents and ambitious interactive and visual work. AWS’s what’s-new note and the Bedrock model card describe it in enterprise language: a frontier model for coding, agentic tasks, and knowledge work, with a focus on long-running agents and more ambitious interactive work.

The specs line up across xAI and the model card:

500K token context window
Configurable reasoning at low, medium, high, and xhigh
Image and text in; text out
No audio, speech, video, or embedding support

That context window is the practical reason to care. A 500K window is large enough to keep a repo, a design pack, traces, and screenshots in one request instead of constantly compacting history. Reasoning effort is a dial, not a separate model: turn it down for cheap classification, turn it to high or xhigh when an agent needs to think through a long task.

$2 in, $6 out

xAI publishes Bedrock token prices as $2 per 1 million input tokens and $6 per 1 million output tokens. AWS’s August 19 what’s-new note does not list dollar amounts. It does say the Global inference profile is offered at a lower per-token cost, without publishing the discount.

US Geo vs Global

AWS is selling residency and capacity as much as the weights.

The US Geo profile, us.xai.grok-4.6, routes requests only inside the United States. AWS says that lets you scale while keeping data processed in the US to meet data-residency requirements.

The Global profile, global.xai.grok-4.6, serves requests from any commercial AWS Region where the model is available. AWS says that gives the broadest access to Bedrock capacity and the highest throughput during demand spikes, at that lower per-token cost.

Cross-Region inference automatically routes requests across Regions so you do not have to manage capacity yourself. AWS says the feature is available in all AWS Regions where Amazon Bedrock is offered. xAI says the model is available to all developers in supported AWS Regions.

APIs and the controls enterprises already use

The model runs on the bedrock-runtime endpoint. Supported APIs are Responses, Chat Completions, and Converse. The model card also lists bedrock-mantle as a supported endpoint. Invoke is not supported.

This is why Bedrock GA is how many enterprises actually adopt a frontier coding and agent model. AWS says Grok 4.6 uses the same account-level controls as other Bedrock models: invocation logging to Amazon S3 or Amazon CloudWatch Logs, CloudWatch metrics, and cost itemization in AWS Cost Explorer and the AWS Cost and Usage Report.

On bedrock-runtime, the model card lists support for projects (default project only), invocation logs, response streaming, prompt caching, and reasoning.

It also lists what that endpoint does not do for Grok 4.6: server-side tool use, intelligent prompt routing, count tokens, structured outputs, and application inference profiles. If you are wiring tools into an agent, plan to do tool dispatch on your side. Bedrock will not run server-side tool use for this model.

SpaceXAI on the label, August 18 on the card

Two documentation mismatches are easy to trip over.

AWS’s what’s-new headline brands the release SpaceXAI Grok 4.6. xAI’s post and the Bedrock model card say xAI — Grok 4.6. The inference profile IDs are the xAI ones: us.xai.grok-4.6 and global.xai.grok-4.6.

The model card lists a launch date of August 18, 2026. Both GA notes are dated August 19, 2026. Lifecycle is Active. End-of-life is listed as N/A.

If you already run models on Bedrock, pick a profile and send a long-context Responses, Chat Completions, or Converse request. Use us.xai.grok-4.6 when US residency is required. Use global.xai.grok-4.6 when you want the widest capacity and AWS’s lower Global token rate. Set reasoning to high or xhigh, turn on streaming and prompt caching, and confirm invocation logs land in the S3 bucket or CloudWatch log group you already use. Start with the Bedrock model card, then the AWS what’s-new note, for profile IDs, API surface, and the feature matrix before you point a production agent at it.