Dedicated Inference

Reserved throughput or dedicated replicas for a specified model, context length and concurrency. Capacity commitments are agreed separately from token usage.