Amazon Replaces Claude Sonnet 4.6 with Strands Decider for Ticket Labeling, Cutting Costs by 98%

·2026.11.09 09:00

Key point

Amazon engineers swapped Claude Sonnet 4.6 for the Strands Decider E4B model in their internal ticketing system, reducing per-ticket costs from $0.0222 to approximately $0.0005.

Details

  • Amazon replaced Claude Sonnet 4.6 with Strands Decider E4B for internal ticket labeling, achieving a 98% cost reduction and 78% latency improvement.
  • The Strands Decider E4B model processes tickets in 897 ms on average compared to 4,215 ms for Sonnet 4.6, using 2,010 total tokens per ticket versus 6,724.
  • Cost per million tickets dropped from approximately $22,200 for Sonnet 4.6 to about $500 for Strands Decider E4B, saving millions in annual token costs.
  • The implementation uses a three-way configuration switch (BEDROCK, DUAL, DECIDER) via AWS AppConfig to allow parallel testing and instant rollback.
  • Strands Decider E4B runs on SageMaker ml.g5.xlarge instances (A10G, 24 GB), utilizing a 4,096-token window with rewritten label criteria to fit context.
  • The decision model approach is suitable for high-volume, closed-choice classification tasks where explanation text is not required and context fits within a few thousand tokens.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.