Note

OpenAI advertises $0.06 per minute for realtime transcription. Sounds cheap, right?

The catch: every new audio chunk often requires sending the previous transcript again to maintain context. That means real costs can be much higher—closer to $1 per minute in practice.

Alternatives like Soniox is $0.12 per hour (30 times cheaper) but streaming services usually have the same context issue.

The key question: which service handles context most efficiently?