OSS
Superwhisper's s1-mini Is a 0.6B Text Normalizer, Not a Transcriber, and It Is Trending on Hugging Face at Only 3,474 Downloads Against 232 Likes
s1-mini is Superwhisper's first open-weights model, fine-tuned from Qwen3-0.6B, and it does not touch audio at all. It takes a raw ASR transcript and rewrites it as clean written text, removing fillers, resolving false starts to whatever the speaker settled on, applying punctuation and capitalization, and rendering spoken numbers, dates, currency and email addresses in written form, hitting 94.8% token accuracy on a held-out set of 7,519 English cases. The quantized build is 462 MiB and runs on a laptop CPU, and its position at trend rank 15 on 3,474 downloads is unusual, meaning it is being noticed rather than deployed.
Source
↳ Follow the thread