faster-whisper
FreeA CTranslate2 Whisper implementation — same models, faster, less VRAM
A Whisper inference stack rebuilt on CTranslate2: same models, faster, lower VRAM, and long the default backend for many projects. A Python library with no GUI. Note that maintenance has stalled, with the last commit in November 2025.
Why we picked it
faster-whisper is genuinely fast: the same models finish sooner and use less VRAM than the reference implementation, and many downstream projects default to it. 24,911 stars, MIT.
But the current state needs stating plainly: maintenance has stalled. The last commit was 19 November 2025 and the last release, v1.2.1, was 31 October 2025 — over nine months of no activity as of August 2026, with 315 issues left open.
The repo is not archived and the code still works, so existing projects are fine. If you are wiring it into something you plan to maintain for years, factor that in — or look at whisper.cpp.