Router: Cutting AI Inference Costs by 40%

Router matches inference requests to the lowest-cost model that still meets performance needs, adjusting in real time to latency and failure rates. Reported result: an average 40% cut in AI inference spend right now.