Language-model endpoints with their cost, latency and quality.
Reference language models with published prices per million tokens, context limits and measured latency, so you can compare and route to the right one. Link a model to a routing target to send it real traffic.
