Superwhisper Open-Sources S1-mini, a 0.6B On-Device ASR Post-Processing Model

On August 20, Superwhisper released S1-mini, its first open-weights language model, which quickly climbed the Hugging Face trending chart. The model has only 0.6B parameters, is built on Qwen3 with a Transformers architecture, and focuses on post-processing ASR transcripts: cleaning spoken-style transcriptions into standardized written text.

Confirmed

• superwhisper/s1-mini is open-weights on Hugging Face and trending; it is Qwen3-based with a Transformers architecture and 0.6B parameters
• Capabilities include text normalization/denormalization, punctuation restoration, casing correction, number and date formatting, and removal of filler words and self-corrections
• Supports 100% local execution, is available for trial in the Superwhisper App, and runs directly in the browser via 🤗 Transformers.js on WebGPU without a server

Why it matters

• Raw ASR transcripts are often cluttered with spoken artifacts; S1-mini offers post-processing at a tiny parameter count, balancing privacy (local inference) with low deployment cost
• The browser WebGPU path lowers integration barriers, letting dictation and meeting-notes apps clean transcripts directly on the web

2026-08-20 ~ 2026-08-20 · 5 related posts

• [source] Superwhisper launches S1-mini, a 0.6B on-device model for transcripts — Scobleizer · 2026-08-20
• Superwhisper releases S1-mini: 0.6B parameter text normalizer for STT output — Recoil42 · 2026-08-20
• [source] S1-mini runs locally in browser for speech text cleaning — xenovatech · 2026-08-20
• Superwhisper releases S1-mini: 0.6B LLM for local transcript cleaning in the browser — pcuenq · 2026-08-20
• [source] Qwen3-Powered ASR Model superwhisper/s1-mini Trends on Hugging Face — superwhisper · 2026-08-20