Grok 4.6 hits AWS Bedrock: 500K context, $2/$6 per M tokens
xAI's flagship agent model is now one Bedrock call away — cheaper than Sonnet, four reasoning tiers, and cross-region routing baked in on day one.
Copy markdown
Grok 4.6 is one Bedrock call away
xAI's long-running agent model landed on Amazon Bedrock on Aug 19 with a 500K-token context and four reasoning tiers (low / medium / high / xhigh). If you're already routing through Bedrock, it's now a model-ID swap.
$2 in, $6 out — cheaper than Sonnet
Grok 4.6 on Bedrock lists at $2 per million input tokens and $6 per million output, with cached input at $0.50/M. That's roughly half of Claude Sonnet 4.5, making xAI's frontier tier the cheap agent pick on AWS.
Cross-region inference, day one
Two inference profiles ship at launch: `us.xai.grok-4.6` keeps requests inside US regions for data residency, and `global.xai.grok-4.6` routes to any commercial AWS region for throughput and cost. Pick your tradeoff without extra wiring.
500K context aimed at long agents
xAI is positioning 4.6 for "long-running agents and ambitious interactive and visual work." 500K tokens is enough for a full repo plus a session's worth of tool-call history in a single prompt — the reason to reach for it over cheaper Bedrock models.
Elsewhere: A2A joins MCP at the Linux Foundation
Google is transferring the Agent2Agent protocol to the Agentic AI Foundation, putting agent-to-agent messaging under the same neutral governance as Anthropic's MCP. Two of the biggest agent standards now sit in one foundation.
Elsewhere: GLM-5.3 shows up on LLM Gateway
Z.ai's coding-tuned MoE (743B, Terminal-Bench 3.0 at 28.3 vs 5.2's 4.6) is now callable through llmgateway.io with a unified key. Public open weights are still held back pending safety review.