Inference Cost Floor

Inference Cost Floor is the per-deliverable token spend below which an agency cannot price a retainer without eroding gross margin. It matters because model pricing moves independently of the value a client perceives: a workflow that costs $0.40 per run today can cost $2.10 after a provider reprices a frontier tier, and the retainer does not move with it. Agencies that track cost per prompt, per report, or per agent run can reprice or reroute before a quarter closes. Routing layers such as OpenRouter and gateways like Helicone or Portkey let a delivery team shift traffic between Anthropic, OpenAI, and open-weight models without rewriting client code. The floor is not a single number; it is a range that shifts with context length, retries, and caching. Teams that measure it monthly keep 20 to 40 points of margin on AI-heavy scopes; teams that do not discover the floor only when a client asks why the invoice grew.