Research
Cactus ships Whistle, a 16.9 MB speech-to-text model that beats Whisper Base on CPU
Cactus released Whistle, a 16.9 MB speech-recognition model that runs on CPU and reports better LibriSpeech results than the 145 MB Whisper Base. Tiny, fast and local is a product category of its own: on-device transcription with no round trip, no per-minute bill and no audio leaving the machine. If voice is on your roadmap and latency or privacy is the objection, this is worth a weekend test.
Companies mentioned
