Does OpenAI API have hidden rate limit?

Recently, on our servers, we consistently get error notifying us we overshot the token limit, but we are like very sure we are far from even cross the threshold.

e.g. our context prompt is less than 1k, but openai is consistently telling us we are beyond 8k. (GPT-4 limit)

When we run the completion, it keep on throwing this error.

Does Open AI have hidden rate limiting mechanism?