Whistle: 16.9 MB on-device speech-to-text model on a public model platform
Cactus-Compute published Whistle, a 16.9 MB speech-to-text model for on-device use. According to the author, it rivals Whisper base for that kind of use.
On October 9, 2026, a Radar briefing reported that Cactus-Compute published Whistle, a 16.9 MB speech-to-text model made available on a public model platform. The author states that it rivals Whisper base for on-device use. The source suggests that a small ASR model running locally could replace Whisper base where size and privacy matter. The briefing gives no accuracy metrics, supported languages, license or evaluation sets, so these points should be treated as the author's claims.
To check the material, open the original model page on the public platform named by the source and read the model card, the license, the languages and the evaluation results published by its authors. Compare the figures with Whisper base on audio similar to yours before adopting any replacement. Rota Nacional does not host or run Whistle. If you use AI to study this material, do not paste recordings or transcripts containing personal data into external tools, and remove identifiers before any query.