xAI's Grok 4.6 is now available in Amazon Bedrock

xAI's Grok 4.6 foundation model is now available through Amazon Bedrock, expanding xAI's presence on AWS's managed model catalog beyond the earlier Grok 4.3 release.

What's new

AWS's Machine Learning Blog announces: "Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work." The model is reachable via cross-Region inference profiles from US, Canada, European, Asia Pacific, Middle Eastern, African, and South American regions, with a separate Bedrock-mantle endpoint limited to us-west-2 in-Region access.

Grok 4.6 on Bedrock ships with a 500K-token context window, four configurable reasoning-effort levels (low, medium, high, xhigh), image input support, prompt caching, and tool calling, accessible through both an OpenAI-compatible API and AWS's native Converse API. Pricing starts at $2.00 per million input tokens for global inference ($2.20 for US-only inference) and $6 per million output tokens, with a Priority tier at 1.75x standard rates and a Flex tier at 0.5x.

Context

AWS has been steadily expanding Bedrock's roster of third-party frontier models throughout 2026, recently adding OpenAI's GPT-6 Astra and Moonshot AI's Kimi K3 to the same managed catalog. Grok 4.6 itself launched on Bedrock's runtime back on August 18, 2026; this announcement formalizes broader regional and endpoint availability. The listing follows xAI's own release of Grok 4.7 as its new frontier flagship, meaning Bedrock customers now have access to a still-recent, one-generation-back Grok model rather than the very latest.

Why it matters

Every major frontier lab now treats multi-cloud distribution as standard practice — xAI landing on Bedrock alongside OpenAI and Moonshot AI gives AWS enterprise customers a growing menu of non-Amazon frontier models without leaving AWS's billing, IAM, and compliance surface. For xAI, Bedrock access extends Grok's reach into AWS's enterprise customer base, which skews toward regulated industries that are often slower to adopt a new vendor's own first-party API directly.