According to CoinMeta, Superwhisper has released its first open-source model, s1-mini. This model has approximately 600 million parameters and is fine-tuned based on qwen3-0.6b. It is specifically designed to organize drafts from speech transcription and can run entirely on local devices. s1-mini itself does not handle audio listening; instead, it relies on speech recognition models such as whisper and parakeet to convert sound into text before further processing. The model removes filler words, corrects mispronunciations, adds punctuation and capitalization, and converts spoken numbers, dates, times, amounts, and email addresses into standard written format. Currently, the model only supports English. The official quantified version occupies about 462MB of space and can run on a regular laptop CPU. It also provides gguf for integration with llama.cpp, ollama, lm, and studio. On 7,519 pieces of English test data, the official team achieved an accuracy rate of 94.8% with this model. s1-mini began experimental testing in June, and its processing speed can reach about 200 token per second.