Strands Harness claims 28% lower agent cost

Strands released an Apache-2.0 general-purpose agent harness that costs 28% less than matched Claude or GPT harnesses across six benchmarks, according to its launch announcement. The same post reports a 77% cost reduction against Claude Code with Fable 5, but both figures are vendor benchmarks rather than independent evidence. Builders considering a deployable agent runtime should test its token savings against their own traces and keep the archive’s coding-agent portability check in the migration plan.