An upper bound for the number of tokens that can be generated for a response, including visible output tokens and reasoning tokens.