vbtm.
verbatim.

I built this because Wispr Flow was the best thing I'd added to my workflow in years — and I wondered whether I could build something just as good.

A few months and one obsessive STT benchmark later, this is what I shipped. We picked Mistral Voxtral Mini — a transcription model that outperforms GPT-4o mini, Gemini 2.5 Flash, AssemblyAI Universal, and Deepgram Nova on accuracy, and runs roughly 3× faster than ElevenLabs' Scribe v2. Then we wired it to a polish step that knows which app you're dictating into, and let you rewrite the AI's instructions yourself if our defaults aren't your style.

Why is it this cheap?

Verbatim is priced almost exactly where it stops making money. Each dictation has a real API cost — speech-to-text plus a small polish call — and Core is set just above that, with enough margin to keep the lights on and one developer fed. Pro adds a few dollars on top for the custom-prompt features, which cost us almost nothing extra to deliver but are worth paying for if you care about voice.

The honest reason it's priced this low: I don't think anyone else would build this at this price. The unit economics only work if you're a tiny team that doesn't have to pay sales reps, VCs, or a marketing agency. If we ever raise prices it'll be because the API costs went up — not because we got bigger.