Cheap Open-Weight Models Don't Cut Agent Costs Without Unit-Economics Routing

A 9:16 walkthrough of the post's argument: a low price per million tokens is not an agent cost model. Routing has to use cost per successful task, escalation rate, GPU utilization, and power. The bar chart uses the post's labelled illustrative numbers, not measured benchmarks. Full write-up: https://hackyjs.com/posts/cheap-open-weight-models-dont-cut-agent-costs-without-unit-economics-routing

inference-economicsllm-routingopen-weight-modelstoken-budgetsai-infrastructure

Read the post →