Grok 4.7 is now available on Amazon Bedrock

xAI’s Grok 4.7 has landed in Amazon Bedrock, putting a new frontier model into AWS for coding, long-running agents, and knowledge work. It comes with a 500K token context window, four reasoning levels, and support for text and image input. The model is served through cross-Region inference profiles, and you can reach it with the Responses API, Chat Completions, or Converse.

The headline here isn’t raw speed. xAI says Grok 4.7 is its most capable model for coding and knowledge work because it stays on difficult tasks longer and checks its own output more carefully before moving on. That self-verification angle matters. On long tasks, the first mistake is usually the one that drags everything else off course.

xAI also says the model was trained on a larger base model with a longer reinforcement learning run over harder tasks, weighted toward jobs that can take many hours. It says that training improved both self-checking and the model’s use of its 500K-token window. The company also points to stronger document and presentation generation, plus gains in work that resembles what lawyers, nurses, and financial analysts do.

Independent testing from Artificial Analysis shows gains too. Its Intelligence Index rose to 46 from 44 on Grok 4.6, while the Coding Agent Index moved to 56 from 47. The AA-Briefcase score for long-horizon knowledge work climbed to 1,657 from 1,546, and GDPval-AA for professional work products increased to 1,695 from 1,605. But there’s a tradeoff: Grok 4.7 used about 81k output tokens per Intelligence Index task, roughly double Grok 4.6’s 38k.

AWS is also packaging the model with the usual Bedrock machinery. You get prompt caching, Guardrails, structured outputs, invocation logging, and the option to route through a US geography profile or a Global one. The practical move is obvious: pick the reasoning level on purpose. Leaving it at the default high setting for every call is how costs and latency quietly get away from you.