Strands Harness cuts agent costs by 28% while matching frontier accuracy

Strands harness is a general-purpose agent harness that runs on Claude, GPT, and other models. Across six benchmarks, it costs 28% less than Claude Code, Codex, and similar tools while scoring equal or better. With Fable 5, it cost 77% less than Claude Code and scored higher on Terminal Bench 2.1. Built-in prompt caching and context management drive the savings. One line of Python or TypeScript gets you started; Apache 2.0 licensed.