An MSP now sells "AI cost management" as a recurring line. This is the page that produces the number the client's CFO signs. Put in the monthly token volumes, get what the workload costs today, what the cheapest capable model would cost, and the annual difference. No login, nothing uploaded, every price dated.
| Model | M tokens in /mo | M tokens out /mo |
|---|
Cost = (millions of input tokens × input price) + (millions of output tokens × output price), with both prices per million tokens, read from the OpenRouter catalogue of 426 listed models on 2026-10-31. The comparison set is every model that can hold the context size in use (1,048,576 tokens). No savings are assumed beyond a like-for-like swap on published prices; nothing here accounts for quality differences, eval results or migration cost, which is the review a human does afterwards.