Grok 4.7 Comes to Amazon Bedrock With a 500,000-Token Window
xAI's Grok 4.7 is now listed in the Amazon Bedrock model catalog, presented as a frontier model built for coding, long-running agents and knowledge work. The concrete details for anyone wiring it into a product: a 500,000-token context window, and reasoning effort you can dial across four settings — low, medium, high and xhigh.
It is served on the bedrock-runtime endpoint through cross-Region inference profiles and supports the Responses, Chat Completions and Converse APIs, which puts it on the same invocation plumbing as the other hosted models rather than behind a separate client. Whether "long-running agents" describes behaviour rather than a marketing category is a question only multi-hour tasks can settle.
AI-generated discussion · Different perspectives
The substantive claim behind the label is about training: a longer reinforcement-learning run over a harder mix of tasks, deliberately weighted toward problems that take many hours to complete.
Those capability and training details come from xAI's own launch announcement, and the line about working longer on difficult tasks while checking its own work more carefully is vendor positioning rather than a benchmark result.
It is still a testable bet rather than pure marketing: tuning on hours-long problems predicts gains on long-horizon agent work, not on one-shot leaderboard questions.