Should I run OpenAI Whisper locally or use the API?
OpenAI Whisper is available two ways: the OpenAI Audio API (cloud-hosted, pay per minute) or open-source weights you run on your own hardware. The right choice depends on volume, privacy, and latency tolerance.
Use the OpenAI API when: you transcribe occasional short clips, want zero setup, accept that audio is uploaded to OpenAI, and your monthly volume stays under roughly 200 minutes. The current API price is 0.006 USD per minute, so 100 minutes per month costs 60 cents.
Run Whisper locally when: you dictate daily, transcribe more than 30 hours per month, work in a regulated industry (medical, legal, finance, government), or simply prefer audio not to leave your device. Local Whisper has zero per-minute cost after install and works offline.
For Windows users, the practical local-Whisper path is StarWhisper. It bundles whisper.cpp (the optimized C++ port of Whisper), handles model downloads, integrates with the Windows hotkey/clipboard system, and types transcribed text into any app. The free tier covers 500 words per day with the Tiny, Base, and Small models. Pro at 10 USD per month gives unlimited use and unlocks Medium and Large models. NVIDIA CUDA acceleration delivers 5 to 10 times speedup, often faster than real time.
Break-even math: at 0.006 USD per minute API cost and roughly 150 spoken words per minute, the API costs about 0.04 USD per 1,000 dictated words. StarWhisper Pro at 10 USD per month breaks even at about 250,000 dictated words per month, which is a few hours of daily dictation. If you are heavier than that, or care about privacy at all, local is the answer.
For Mac, similar local-Whisper options include Superwhisper and MacWhisper. For Linux, the whisper.cpp CLI directly. For Windows, StarWhisper covers both the dictation flow and the Whisper engine in one install.
All FAQ topics