Amazon Bedrock Adds xAI's Grok 4.6 Model

Amazon Web Services has added xAI's flagship Grok 4.6 model to Amazon Bedrock under the ID xai.grok-4.6. Released on August 19, 2026, one week after its debut, the model costs $2 per million input tokens and $6 per million output tokens directly. On Bedrock, In-Region and Geo cross-Region requests are priced at $2.20 per million input and $6.60 per million output tokens, while Global cross-Region inference is $2.00 and $6.00. Cached-read pricing is $0.55 per million tokens for In-Region/Geo and $0.50 for Global. The model features a 500,000-token context window, accepts text and image inputs, and offers four reasoning tiers: low, medium, high, and xhigh.

This release expands on Grok 4.3, which debuted on Bedrock in June 2026 at $1.25 per million input and $2.50 per million output tokens with a 1-million-token context window. Unlike its predecessor, which only ran on the bedrock-mantle endpoint, Grok 4.6 also utilizes the bedrock-runtime endpoint. This allows deployment across more than 30 AWS Regions via Geo and Global cross-Region inference profiles, though the mantle endpoint still serves In-Region requests from us-west-2 (Oregon).

Grok 4.6 shows strong benchmark gains, scoring 61 on the Artificial Analysis Intelligence Index, up from 56 for Grok 4.5 High, matching GPT-5.6 Sol Max and trailing Fable 5 Max by one point. On agentic suites, it scores 65.9% on DeepSWE v1.1, 57.5% on APEX-Agents, and 26% on Terminal-Bench v3.0, compared to Grok 4.5's respective scores of 54%, 47.1%, and 15.7%.

Developers can access the model via the OpenAI SDK or AWS's Converse API. While prompt caching works on both endpoints, server-side tool use is unsupported on the runtime endpoint, whereas the mantle endpoint supports client-side tool calling and abuse detection. This integration allows enterprises to run Grok 4.6 within their existing AWS security and compliance perimeters.