This is what MIT says about a similar issue:
"The quota code maintains three items for each account. A current usage, the account quota and the last time the quota record was updated.
When you do a new query, I check your quota usage. I do this by calculating:
quota * (current_time - timestamp)
new_usage = current_usage - ----------------------------------
seconds_in_a_day
If new_usage is less than quota, access is permitted. Otherwise, an over-quota error is sent. Also, new_usage is set to zero if it is less than zero. The values of timestamp and usage are then updated.
If you issue a prompt that uses a lot of tokens, then it can take a while for your computed usage to drop back below quota.
Keep in mind that ChatGPT maintains no state on its own. When you have a continuing conversation with it, each time you add a new prompt, the whole conversation is played back to ChatGPT. This can add up to a lot of tokens very quickly!
Unless you use the "Reset Conversation" function, then each use of the "Converse" block keeps adding to the conversation (and the token count)."
You are using the default API settings. You probably could eliminate this issue by signing up with Chatbot for an api key.