Releasing and pricing an AI model with tenfold capability at tenfold cost

A new model is ten times as capable as the one your product uses today, and it also costs ten times as much. What do you do? ```hint Two tens do not cancel Ask whether the extra capability is worth the same amount on every request, and set that against a cost that applies to every request the new model serves. ``` ### Clarifying Questions - Does "ten times the cost" mean the cost to serve each request, the cost to train the model, or the price customers would be charged? - How is "ten times as capable" measured: benchmark scores, the share of real user tasks it completes that the current model fails, or abilities the current model lacks entirely? - Who would get it: consumers on a subscription, developers paying per token through an API, enterprise customers on contracts, or all three? - Is the decision whether to release it at all, how to price and package it, or whether it should replace the current model? ### What a Strong Answer Covers - What the tenfold capability and the tenfold cost each mean, clarified before any recommendation - Unit economics per request and per user, with labeled assumptions and a break-even condition - Which users and tasks gain enough from the extra capability to justify its cost - Release, packaging and pricing options weighed against each other, plus levers that lower the cost over time, ending in a clear recommendation - Metrics and guardrails that would confirm or reverse the decision after launch ### Follow-up Questions - Usage of the new model grows much faster than the revenue it brings in. What do you change? - A competitor releases a model of similar capability at a third of your price. How does your plan change? - How would you decide when to retire the cheaper model?