xAI's Grok 4.6 is now available in Amazon Bedrock

Amazon has added xAI's Grok 4.6 to the Amazon Bedrock model catalog, the company said on its Machine Learning Blog. The model, which xAI built for long-running agents, coding and knowledge work, went live on Bedrock on August 18, 2026, with a 500,000-token context window and four selectable reasoning effort levels: low, medium, high and xhigh.

A second xAI model, now on two endpoints

Grok 4.6 is xAI's second model in Bedrock. Grok 4.3, the first, was reachable only through Bedrock Mantle, the OpenAI-compatible inference engine in Amazon Bedrock. Grok 4.6 runs on both the bedrock-mantle and bedrock-runtime endpoints and supports the Converse API alongside the Chat Completions and Responses APIs; the Invoke API is not supported. It accepts text and image input and returns text: audio, video and embedding modalities are absent, and it does not generate images.

What is new on Bedrock

Several capabilities arrive with this launch rather than carrying over from Grok 4.3. On bedrock-runtime, converse and converse_stream are both available, so streaming uses standard Converse events, and reasoning effort gains a new xhigh level. Cross-Region inference replaces in-Region inference there: us.xai.grok-4.6 keeps traffic inside the US geography for data-residency requirements, while global.xai.grok-4.6 routes worldwide and costs $2.00 per million input tokens against $2.20. Grok 4.6 also supports Bedrock Guardrails on bedrock-runtime — content filters, denied topics, PII redaction and word policies — plus invocation logging in CloudWatch and prompt caching at roughly a quarter of the standard input rate.

Benchmarks, regions and pricing

xAI's published figures for Grok 4.6 High, released on August 12, 2026, include an AA Intelligence Index score of 61, GDPVal-AA v2 of 1753, 69.9% on CursorBench v3.2, 65.9% on DeepSWE v1.1 and 57.5% on APEX-Agents. Availability differs by endpoint: bedrock-mantle runs in-Region in US West (Oregon), while bedrock-runtime is reachable through Geo and Global cross-Region profiles. Standard, Priority (1.75x) and Flex (0.5x) tiers are offered, with pricing from $2 per million input tokens and $6 per million output tokens.

Structured outputs, server-side tool use, intelligent prompt routing and count tokens work only on bedrock-mantle, while invocation logging and the Converse API point to bedrock-runtime.