xAI’s Grok 4.6 is now available in Amazon Bedrock
Amazon Web Services Suheel Farooq ● Covered by 9 sources
xAI’s Grok 4.6 is now in Amazon Bedrock. It brings a 500K-token model for long agent jobs, plus new Bedrock endpoints and streaming.
Based on reporting by Amazon Web Services, Suheel Farooq — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AWS says xAI’s Grok 4.6 is now available in Amazon Bedrock, and this one is aimed squarely at the awkward, expensive stuff: long-running agents, coding, and knowledge work. It launched on Bedrock on August 18, 2026, comes with a 500K token context window, and lets callers dial reasoning effort to low, medium, high, or xhigh.
This is xAI’s second model in Bedrock, but the surface area is much wider than the earlier Grok release. Grok 4.6 runs on both bedrock-mantle and bedrock-runtime, and it supports the Converse API alongside Chat Completions and Responses. That matters because Bedrock users can now reach it through the standard runtime path, not just the OpenAI-compatible Mantle route.
The model also picks up a few things that make it more interesting for agent builders. On Bedrock runtime, AWS is exposing cross-Region inference profiles instead of in-Region hosting: us.xai.grok-4.6 for US geography and global.xai.grok-4.6 for worldwide routing. Bedrock Guardrails work here too, with content filters, denied topics, PII redaction, and word policies evaluated against both prompt and response. Invocation logging is supported as well, so runs can land in CloudWatch with the request, response, token counts, and the inference profile used.
There are limits, though. Grok 4.6 accepts text and image input and returns text, but not audio, speech, video, embeddings, or image generation. Structured outputs are available on bedrock-mantle, not bedrock-runtime. And if you care about cost, the runtime routing and service tiers matter: Standard is the default, Priority costs 1.75x, and Flex cuts to half price. AWS is clearly pushing this as a serious production model, not a toy demo with a flattering blog screenshot.
The launch post also leans hard on xAI’s own claims about the model’s behavior: stronger self-checking on long trajectories, better first passes on visual and interactive work, and improved safeguards backed by pre-deployment, post-deployment, and third-party testing. Benchmark numbers are there too, but the real story is simpler: Bedrock just got a more capable Grok, and AWS is giving it the full enterprise plumbing.
My take — AI-written commentary, not fact-checked reporting
AWS keeps doing the thing it does best: turning model launches into procurement artifacts. That’s not a diss, it’s the business. The interesting bit is that Grok 4.6 arrives with enough knobs, guardrails, and logging to look less like a chatbot and more like something a compliance team can tolerate while the agents quietly chew through the work.
Read more about this at: Amazon Web Services
