Amazon saves millions by replacing Sonnet 4.6 with Strands Decider

Amazon engineers replaced Claude Sonnet 4.6 with Strands Decider to label internal support tickets, maintaining matching performance while reducing costs by 98%. The decision model processes closed categories much faster and with significantly fewer tokens by returning a probability distribution rather than generated text. The team implemented a three-way configuration switch in AWS AppConfig to test both models in parallel on live traffic before fully cutting over. Open article →