openai/whisper

Choose this fork only if batch transcription is the main requirement and you can tolerate staleness. For general Whisper use, upstream is the safer default.

Choose this fork if your goal is deployment and API shaping, especially on Cog, and you value segment output and extra request controls. Choose upstream if you want the actively maintained, broader Whisper project with newer fixes, docs, notebooks, and model support.

Choose this fork if low VRAM is the primary blocker and you can tolerate being far behind upstream. If you want the latest Whisper fixes, compatibility work, and general-purpose maintenance, upstream is the safer choice.

Choose this fork if your priority is an interactive web UI and simple device choice. Choose upstream if you need current Whisper features, ongoing maintenance, or a scriptable Python library.

Prefer this fork if inference efficiency is the main goal and you are willing to own maintenance. Prefer upstream if you want current fixes, broader compatibility, and lower risk.

Prefer upstream unless you explicitly need this older snapshot. The fork adds no visible capabilities, while it lags substantially behind upstream maintenance and fixes.

Choose upstream unless you specifically need this fork’s namespace or ownership. There are no concrete added features here, and it is one commit behind on a CI-related maintenance fix.

Choose this fork if you want a WebRTC/web-app transcription prototype and are comfortable owning maintenance. Choose upstream if you need the actively maintained Python library, CLI, notebooks, and broad compatibility with less operational overhead.

Prefer upstream unless you specifically need an old, frozen Whisper snapshot; this fork adds no visible product value and is materially behind current upstream.