Strands Decider replaces Sonnet, slashes Amazon costs

This is a compelling case for using specialized decision models instead of expensive general-purpose LLMs when the task is tightly bounded.

• –Cost fell from roughly $22,200 to $500 per million tickets.
• –The approach works because tickets map to a fixed catalog of 20 labels and require a choice, not an explanation.
• –Strands Decider reduced output to a single decision, cutting total tokens per ticket by 70%.
• –Amazon’s BEDROCK/DUAL/DECIDER rollout switch enabled live comparison and a quick rollback.
• –The result does not generalize to open-ended reasoning, rapidly changing taxonomies, or workflows requiring rich explanations; traditional classifiers may also be viable alternatives.