Hi, everyone!
First off, thank you all for the thorough feedback. Rest assured, we’ve passed every single one to the v0 team.
Our inference costs are based on input and output tokens. Trying to normalize to a single “token” or “message” shared across input and output is very difficult based on the shape of our usage. To get around that, our competitors are just jacking up their pricing to .25c+ per message regardless of the request and response size (as a comparison, the median message on v0 today cost $.08 – more than 3x less). Those inflated per-message prices seem unfair to us.
We decided the most transparent, future-proof, reliable solution is to take inspiration from how frontier model labs charge for on-demand inference… with a credit pool and incremental burndown based on consumption. That’s what we’ve done here.
We will introduce new models with different prices in the coming weeks, including the ability for you to easily switch between cheaper and more expensive options. Our new pricing allows us to charge for what you use independent of which model you use – and to support more use cases, like the v0 API we released yesterday.
Let me know if you have any more questions or comments 