Overcoming Unresponsive Support and Rate Limit Issues

API
, , ,
1

Hello OpenAI Community,

We are a language learning startup using GPT and Whisper to provide real-time interactive user experiences. With tens of thousands of daily user sessions, managing our rate limits has become a critical challenge. We stumbled across the default Whisper rate limit: 50 RPM and desperately need an increase to reflect our growing traffic.

Our numerous efforts to contact OpenAI for support include:

  1. Multiple email attempts to support@openai.com

  2. Over 15 submissions of rate limit increase forms, providing varied supporting evidence

  3. Engaging the help chatbot from the OpenAI console

  4. Attempting to reach the sales team

Unfortunately, our requests over the past 3 months have gone unanswered.

To cope with these limitations, we have had to intentionally reduce our traffic. This constraint is now directly impacting our growth, with a resulting revenue loss. As we anticipate significant user growth in the upcoming months, the current situation is not sustainable, and it’s becoming increasingly difficult to rely on OpenAI as a partner due to unresponsive support.

We kindly request your help – if you have previously experienced similar support issues, please share any advice or avenues for successfully reaching OpenAI support, at least getting any sane answer.

Thank you!

2 Likes
2

Hi - did you find a solution? I am building an app on Open AI Whisper, and wondering if there are options to eventually keep up with growth as we exceed this 50 limit.

3

Hi, yes, it turned out that OpenAI open-sourced Whisper, and one can deploy it on-prem (e.g. see HuggingFace) and scale infinitely

4

Are you using dedicated hosting with a GPU? I guess in the long-term it will become more cost effective to host it yourself? I was looking at it and it seems at the moment, OpenAI is more efficient solution for us.

5

For personal use it is often perceived as lower cost to host something yourself as you tend to write off the cost of the power used and the hardware you purchased. This is fine if you are inferencing for just you, but if you start to use your system with a wider customer base then in almost every case it is much more expensive to self host that to take a slice from an existing, at scale, API service, assuming that service is somewhat competitively priced, which does seem to be the case with GPT3.5 and 4, the self hosting method for an Open Source model is an excellent option for R&D and personal use, it is not good commercially.