Meet S1-mini: Superwhisper’s 462 MB Open-Weights Text Normalizer That Turns Raw ASR Transcripts Into Clean Written Text
Superwhisper has released the S1 family of models: S1-Voice, S1-Language, and S1-mini. S1-Voice is a cloud speech-to-text model, and S1-Language is a cloud instruction-following model for cleanup and formatting. The one that is quite interesting outside the app is S1-mini, released with open weights on Hugging Face. S1-mini is a 0.6B text normalizer, not a transcriber and not a chat model. It sits after automatic speech recognition and rewrites raw transcripts as clean written text: fillers rem
Make your prediction
Will S1-mini by Superwhisper appear on the Hugging Face open ASR leaderboard by September 5?
Resolves by Sep 5, 2026
Your prediction
50% · 50/50 coin flip
NOYES
Get smart on it
S1-mini is a small text-cleaning model that improves the output of automatic speech recognition systems by removing filler words, fixing self-corrections, adding proper punctuation and capitalization, and converting spoken forms like "support at example dot com" into written email addresses. The model is released with open weights and is small enough to run on a laptop CPU, making it deployable in various settings from individual apps to enterprise networks where audio transcripts cannot leave internal systems. It works by taking a raw transcript and applying one of several preset combinations across three control axes: styling tone, structure format, and context type. The model achieved 94.8 percent token accuracy on its internal test set and was designed specifically to clean transcripts without adding content, correcting facts, or changing dialects.
Related news
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
Better embeddings help AI systems understand meaning more accurately, improving search, recommendation, and classification tasks across applications.
IBM's new Granite 4.2 models ride the wave of interest in local LLMs
IBM has rolled out the newest models in its family of open-weight large language models designed to be downloaded and self-hosted. The newly launched Granite 4.2 comes in 3B, 8B, and 30B parameter variants. Like previous versions, IBM is taking a decoder-only approach here. These new releases offer a 128,000-token context window natively. The 8B and 30B variants (not the 3B one) also go through an agentic reinforcement learning block; they were trained for expanded capabilities like using the t
