Here I explained the pricing structure to another fellow user. Including MAX etc.
For using AI, if you are not experienced in prompting you may have better success with a more seasoned and stable model like Claude 3.5 which is well capable of coding.
3.7 tends often to be too eager to do things and if you do not use properly rules, prompting and not familiar with AI model differences and how to use each differently you may use up the requests faster.
I tried the better models incl. Max, but my setup with 3.5 works so well that even 3.7 or thinking only is needed for very complex things or special use cases.