Smarter to scale with Whisper API or host whisper yourself?

I’m wondering what is more cost-effective as one scales… to just use the Whisper API, or to host on a GCP VM or Azure VM - what ends up being less?

Math and statistics modeling will answer this for you.

The key piece of information: OpenAI Whisper AI costs nothing if you don’t use it. There is no operation tax except for expiring credits. No power bill even.

1 Like

Okay heres maybe a better reframe…

I have an app that transcribes 45-hour and a half piece of audio. To my knowledge, that will be around 30-40 cents per run. I’m wondering if it makes more sense to just pay a VM and run whisper for free, or to bite that cost.

“Pay a VM” isn’t necessarily what you’d want to do; you’d have to reserve GPU instances at an ongoing cost on most providers.

1000 seconds = 16:40 = $0.10 on OpenAI.

There’s other solutions, made completely opaque because they bill you hourly compute rates with no clarity of that compute capability, besides ongoing cost.

OpenAI has an advantage: keeping servers near 100% with thousands of customers spread over them.