Amazon Bedrock’s Advanced Prompt Optimization now includes an optional prompt‑shortening mode. When enabled, the service tries to produce a shorter prompt that scores at least as well as the original on your custom evaluation. If the shortened version is longer or scores lower, the original prompt is kept. The feature works with any model on Bedrock and any evaluation method, and it shows a compression ratio in the job results.
Why it matters
Developers can lower input‑token costs and latency for large‑scale applications without sacrificing prompt quality.
