AWS adds xAI Grok 4.6 to Amazon Bedrock

AWS has added xAI’s Grok 4.6 to Amazon Bedrock, giving enterprise customers another frontier model option for agents, coding and knowledge-work systems inside AWS infrastructure.

The AWS post says Grok 4.6 has a 500K-token context window and four reasoning effort levels: low, medium, high and xhigh. AWS describes it as built for long-running agents, codebase work, research, visual and interactive projects, and other multi-step tasks where the model needs to keep checking its own work.

The important Bedrock change is not just that another model appears in the catalog. Grok 4.6 is available through both bedrock-mantle, AWS’s OpenAI-compatible inference layer, and bedrock-runtime, which means teams can use the standard Bedrock Converse API, streaming, invocation logging and Guardrails around the model. AWS also supports cross-Region inference profiles for U.S. residency or global capacity.

For builders, that makes Grok 4.6 easier to test next to Claude, Llama, Nova and other Bedrock models without moving agent infrastructure out of AWS. AWS says xAI lists pricing from 2 USD per million input tokens and 6 USD per million output tokens, though customers still need to check current Bedrock pricing and Region availability before production use.