Released August 2026 as Superwhisper's first open-weights model, S1-mini is a Qwen3-0.6B fine-tune (596M unique parameters) trained for one job: inverse text normalization and truecasing of ASR output. It is not a speech recognizer. Audio still goes through Whisper, Parakeet, or similar; S1-mini then removes fillers, resolves self-corrections, and renders spoken numbers, dates, currency, and emails in written form. v1 is English-only, Apache 2.0 plus a naming clause, and ships GGUF builds for llama.cpp.