Advice on whether to use local model or API?

[consultant2b]

Hi everyone,

I’ve been using this app for a few weeks now with my OpenAI API, and I absolutely love it so far. However, I’m curious about the differences in speed and accuracy between running a local Whisper model versus the API. From what I gather, the API currently only offers “Whisper 1,” which I assume is the first official release?

Meanwhile, I understand we are already at Whisper 3.0 Turbo - https://medium.com/@bnjmn_marie/whisper-large-v3-turbo-as-good-as-large-v2-but-6x-faster-97f0803fa933

• Does the API really still use something equivalent to the first Whisper model?
• Is there any tangible advantage (speed or accuracy) in using a newer open-source model (like Whisper 3.0 Turbo) on my own machine?
• If I do run Whisper locally on a personal laptop with 16 GB RAM and just an integrated GPU, which model size would be best for English-only use to find the best trade-off between speed and accuracy?

Thanks in advance for your suggestions and feedback.