inference cost.

Noorain Fathima

LLM Token Pricing Is a Facilities Question

Output costs five times input because decode is memory-bound. Batch costs half because utilisation is the provider's central problem. Beneath both sits a floor made of power contracts and grid queues.