AWS tests Grok 4.7 tool-calling agent on Amazon Bedrock across reasoning effort levels
Overview
AWS Machine Learning Blog reports that a tool-calling agent built on Grok 4.7 answered 67 of 72 test runs correctly on Amazon Bedrock, with all 18 runs at the xhigh reasoning effort level correct.
The low setting answered 15 of 18 runs correctly while using about one-eighth of xhigh's output tokens. The tasks were three search-based questions, six runs each per level.
AWS says reasoning tokens count as output tokens and use up max_completion_tokens before the answer is written, so the limit must leave room for both. Grok 4.7 is reached through the bedrock-runtime endpoint and a cross-Region inference profile.
Written by AI from the articles below · updated Oct 13, 12:14 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- AWS Machine Learning BlogOfficialPickGrok 4.7 on Amazon Bedrock: reasoning effort vs. agent accuracy
AIAWS tested Grok 4.7 on Amazon Bedrock with a tool-calling agent on three search tasks, six runs per task at each reasoning effort level. The agent answered 67 of 72 runs correctly, and all 18 runs at xhigh, while low answered 15 of 18 using about an eighth of xhigh's output tokens.
Heat trend
Not enough continuous observations to show a trend yet.