What AWS announced

AWS presented Strands Harness as a runtime environment designed to make AI agent workloads cheaper to operate by cutting token consumption. According to AWS's official Strands Agents blog, the company benchmarked the harness against an existing execution setup and measured a 28% reduction in token costs.

The 45% figure and how the comparison was framed

The reported savings depend on how the comparison group is defined. Alongside the 28% figure, AWS's materials include a 45% reduction under a different benchmarking condition. This means the headline number is conditional on the definition of the baseline agent setup against which Strands Harness was measured, rather than a single, universally applicable saving.

Independent verification not yet established

The cost-reduction figures so far rest on AWS's own benchmarks, published on the company's official project pages. Independent testing of Strands Harness has not yet been established, so the 28% and 45% numbers should be read as vendor-reported results.

AWS's announcement materials, including the official release card and benchmark data embedded in the Strands Agents blog, are publicly available for review.

Given that the 28% token-cost saving comes from AWS's own benchmarks, varies with the definition of the comparison group, and has not yet been independently validated, the claim is Mostly True.

Verdict: Mostly True