Superwhisper Releases S1-mini, a 462MB Open-Weights Text Normalizer

Superwhisper has released the S1 family of models: S1-Voice, S1-Language, and S1-mini. [1] S1-Voice is a cloud speech-to-text model, and S1-Language is a cloud instruction-following model for cleanup and formatting. [1] S1-mini is published on Hugging Face under Apache 2.0 plus a naming clause, while S1-Voice and S1-Language are hosted services. [1]

S1-mini is a 0.6B text normalizer, not a transcriber and not a chat model. [1] It sits after automatic speech recognition and rewrites raw transcripts as clean written text: fillers removed, self-corrections resolved to what the speaker landed on, punctuation and capitalization applied, and spoken numbers, dates, currency, and email addresses rendered in written form. [1] It is fine-tuned from Qwen/Qwen3-0.6B, covers English only in release v1, and is steered entirely by a three-axis control line placed above the transcript. [1] Superwhisper reports 94.8% token accuracy on a held-out set of 7,519 cases, measured greedy on the quantized build. [1] The Q4_K_M GGUF build is a 462 MB file that runs on a laptop CPU. [1]

The release of S1-mini provides developers with a lightweight, locally runnable option for cleaning up raw speech transcripts, rather than relying on heavy, general-purpose LLMs. By fine-tuning a small 0.6B parameter model specifically for text normalization, Superwhisper has created a highly focused tool. Its small size (462 MB) makes it feasible to embed directly within desktop applications or run securely within an enterprise VPC. This contrasts with their cloud-based S1-Language model, which requires data to leave the user's network. What the sources don't address: how S1-mini's performance compares to using established open-source transcribers like OpenAI's Whisper directly, or if there are plans to expand beyond English in future versions.