Does someone have some intuition regarding the output token count of o1-preview compared to gpt4o?
For instance, for the same input prompt, how many more tokens will o1-preview have than gpt4o? 2x more?
Welcome to the forum! 
Are you using API or ChatGPT?
Cost can be alarmingly higher, especially in the case of a small answer to a complex problem stated as a simple output.
The model also likes to write at great length, for example, assuming that your code snippet refactor needs an entire code base written for it, example usage, a procedural walk-through of operations, and tutorials of how to install Python…
They also bill tokens at a higher rate, despite no evidence a more computationally-expensive AI is actually used (although longer context inferences grow in cost).
Token consumption depends on how long the AI continues “reasoning” internally with multiple unseen model calls based on unseen instructions and texts, building an internal chat with itself.
So 2x is an underestimate, unless you have a task like “write a long story” that has little to think about.
See my price analysis at the end of this reply, a simple text completion task, to see my authority to say “alarming costs”…
TY @_j i was unsure where to put it so I moved it to API with gpt tags
I was informed by ChatGPT users not normally being concerned with the cost of internal reasoning tokens billed as output…
In ChatGPT Plus, you can have your one well-articulated message kick off a reasoning session about multiple problems and multiple or extended outputs to be produced, for maximization of the opportunity to use the model.
(The same applies to your API-developed product vs competitors: people will find the cheapest path to maximum expense for the operator for their particular usage…)
thanks, I will have a look!