To my surprise, I found that errored replies count towards the GPT-4 limit. I tried to analyse a text, and three times I had errors. Specifically, 7/7, 9/9 and 8/8 times.
The error was along the lines of
There was an error generating a response - Regenerate response
Something went wrong. If this issue persists please contact us through our help center at help.openai.com.
The last time, I got the error that:
You've reached the current usage cap for GPT-4. You can continue with the default model now, or try again after 7:14 PM. [Learn more]
Surely, errored replies should not count towards this limit?