Get Wispr Flow quality for $0.30/month
Get Wispr Flow quality for $0.30/month
Wispr Flow is a great dictation app, but it costs about $15 a month. Here is how we get the same quality for roughly $0.30, using OpenWhispr and our own OpenAI key.
Key insights
GPT-4o Mini Transcribe bills per audio minute, not per word or seat.
Two hours of monthly speech costs roughly $0.20 to $0.35 via the API.
Filling the dictionary beats AI cleanup for fixing the mistakes that matter.
OpenAI's API does not train on submitted audio by default.
Switching to a local model in OpenWhispr takes seconds for sensitive calls.
In this post:
Section
Wispr Flow works great, but it locks me into roughly $15 a month for what is basically cloud transcription. Here is the setup I landed on that does the same thing for pennies.
The setup
The app: I use OpenWhispr, a free, open-source, cross-platform dictation app. No subscription, no per-seat fee. It runs on Windows, macOS and Linux and supports both local models and cloud providers.
The transcription engine: instead of OpenWhispr's own capped free cloud or the slower local models, I plug in my own OpenAI API key under Cloud Providers and pick GPT-4o Mini Transcribe. That one change gives me full cloud speed with no monthly cap.
Cleanup: OpenWhispr's free cloud can polish the text if you want it, so it adds nothing to the bill. More on why I turned it off below.
Why it costs almost nothing
OpenAI bills transcription per audio minute, not per word. At roughly 16,000 words a month (about two hours of speech) and GPT-4o Mini Transcribe at $0.003 per minute, my bill lands between $0.20 and $0.35. Even if I upgraded to the full GPT-4o Transcribe, I would still pay only around $0.70 a month at that volume.
Model | Price per minute | Est. monthly cost (2 hrs audio) |
|---|---|---|
GPT-4o Mini Transcribe | $0.003 | ~$0.20 to $0.35 |
GPT-4o Transcribe | Higher tier | ~$0.70 |
Wispr Flow subscription | Flat fee | ~$15.00 |
Why it is as good as Wispr Flow
Wispr Flow runs the same class of cloud speech models on fast servers. Pointing OpenWhispr at OpenAI directly gives me the same model quality and the same cloud speed. I just pay for actual usage instead of a flat monthly fee. Apps and services like TurboScribe feel fast because they run Whisper-class models on cloud GPUs, which is exactly what the API gives me.
For everyday dictation the quality gap between Wispr Flow and this setup is negligible. The price gap is about 50x.
Skip cleanup, fill the dictionary instead
I do not use the AI text cleanup for this. The single biggest quality jump comes from filling OpenWhispr's custom dictionary with my branded words, product names, company names and technical terms. That is what fixes the mistakes that actually matter in my workflow. I turn text cleanup off and spend the two minutes on the dictionary instead.
It is the same mindset behind the free image compression tools I use: cheap, targeted choices beat expensive defaults.
The one trade-off
My audio is sent to OpenAI. Their API does not train on submitted audio by default, but the data does leave my device. For sensitive client meetings I switch OpenWhispr back to a local model. For everyday dictation, the API route is the sweet spot between speed, accuracy and cost.
The net result
~$0.30 a month versus $15 a month. The same experience. About 50 times cheaper. If I dictate more, the math still holds, because the API scales with usage, not with seat count.
Author
Alexander Winter
Founder & Writer
Alexander is the founder of HIVER, writing about design, branding and building products that people love to use. He is deeply rooted in SaaS, tech marketing and branding.
