xAI’s Grok 4.6 is available on Amazon Bedrock with a 500,000-token context window and four reasoning-effort settings: low, medium, high and xhigh. It is the second xAI model on the service after Grok 4.3.
Developers can call it through Bedrock Mantle, Amazon’s OpenAI-compatible interface, or the standard Bedrock Runtime. It also supports the Converse API, Chat Completions, Responses and cross-region inference. That wider interface coverage makes it easier to place the model into existing Bedrock applications.
xAI positions Grok 4.6 for long-running research, coding and knowledge-work agents. The company says training emphasized multi-step tasks and that the model increasingly checks its own work on longer trajectories. It also targets first-pass generation of visual and interactive applications. These capability and benchmark claims come from xAI.
The model first launched on Bedrock on August 18, despite the detailed AWS post appearing later. Teams considering it should therefore distinguish availability from the new documentation, and benchmark the four reasoning levels against their own latency and cost requirements rather than assume that the highest setting is always best.