SpaceXAI's Grok 4.7 becomes available on Amazon Bedrock, with AWS tutorial on reasoning-effort trade-offs for agents

Grok 4.7 from SpaceXAI is now available on Amazon Bedrock through two cross-Region inference profiles (us.xai.grok-4.7 and global.xai.grok-4.7). It supports the OpenAI-compatible Chat Completions and Responses APIs and the Converse API. It offers four reasoning-effort settings (low, medium, high, xhigh), tool calling, structured output and a 500K context window. An AWS blog post also reports a 72-run agent test in which 67 runs were correct. xhigh got 18 of 18 and low got 15 of 18 using about one eighth of the output tokens.

The main change is distribution. Teams already building on Bedrock can now call Grok 4.7 without a separate xAI contract, with AWS billing and governance. The reasoning-effort setting is a practical cost lever, since a low setting that gets 15 of 18 right at a fraction of the tokens may suit simpler agent steps. The change is incremental, though. Grok 4.7 already existed, and this adds a route to it rather than a new capability, so it mainly matters to Bedrock customers weighing model choice and cost.

The 72-run test is far too small to support firm conclusions. The gap between low (15 of 18) and xhigh (18 of 18) is three runs, and the task design and error bars are not shown. The claims about long-horizon work and self-verification come from SpaceXAI's own material, and the post is vendor-authored tutorial content. It tells us nothing about how Grok 4.7 compares with rival models on Bedrock or about real-world cost.

1. 01Bedrock per-token pricing for Grok 4.7 against comparable models, including how reasoning tokens are billed at each effort level
2. 02Independent evaluations on larger agent task sets that show whether the low-effort accuracy and token savings hold up beyond 72 runs
3. 03Evidence of Bedrock customer uptake, such as production case studies, regional expansion beyond the cross-Region profiles, or quota and latency reports from developers