AWS details Grok 4.6 access through Amazon Bedrock

AWS has detailed how customers can access xAI’s Grok 4.6 through Amazon Bedrock, giving AWS customers a managed route to the model through two Bedrock endpoints. The September 21 post describes the service integration rather than a new model launch: xAI introduced Grok 4.6 on August 12, and AWS records its Bedrock launch date as August 18.

Grok 4.6 is available through the bedrock-mantle and bedrock-runtime endpoints. AWS lists three supported interfaces: Converse, Chat Completions and Responses. Converse runs through bedrock-runtime; Bedrock also provides the OpenAI-compatible Chat Completions and Responses interfaces.

The endpoints expose different Bedrock features. AWS says bedrock-mantle includes structured outputs and client-side tool calling. The bedrock-runtime option includes Converse, invocation logging and cross-Region inference support. AWS’s model documentation says runtime access uses US and global cross-Region inference profiles rather than in-Region inference.

AWS and xAI describe Grok 4.6 as supporting a 500,000-token context window and four reasoning-effort settings: low, medium, high and xhigh. AWS documents text and image input with text output. Its model card says the Bedrock integration does not support audio, speech, video, embedding output or image generation.

The Bedrock route is one of several distribution channels opened since Grok 4.6’s August 12 launch. xAI said on August 19 that the model was generally available on Amazon Bedrock. Snowflake separately added Grok 4.6 to Cortex AI in public preview through Cortex AI Functions and Cortex Inference on September 3.

AWS’s materials do not agree on whether the Invoke API is supported. The Grok 4.6 model-card summary presents Invoke as supported, while the September 21 announcement and AWS’s API compatibility table say it is not supported. The discrepancy remains unresolved in the published documentation.