How FluidVoice launched Fluid Speech Pro on X
Sub-1B speech-to-text model for on-device dictation, claiming better accuracy than rivals
What FluidVoice launched
How LaunchVideo.com rated the launch
LaunchVideo.com rating
About the launch
FluidVoice has released Fluid Speech Pro, a 0.6 billion parameter speech-to-text model for on-device dictation on a laptop or phone. It is for people who dictate into email, chat, docs, and code, and for developers who want open weights they can run in their own apps. The model is already live in FluidVoice, a free GPLv3 dictation app for Mac, with Windows, iOS, and Linux builds in beta, and the weights are on Hugging Face under CC BY 4.0 for commercial and non-commercial use. On Hugging Face's Open ASR Leaderboard at the November 2026 launch, FluidVoice reports a 6.06% average word error rate, first among open models at 1 billion parameters and under, and second among all open models. With paid models included, that same post ranks it 10th of 66. The company says the 0.6B model beat open models up to 50 times its size, with 10% fewer errors than the next open model under 1B, and that it was the only open model under 1B below 10% word error on the noisy URGENT 2024 set (9.83%). It auto-detects 25 languages, English plus 24 other European languages, and FluidVoice reports 20% fewer errors than its base model on those non-English languages. Fluid Speech Pro uses a FastConformer encoder and a TDT decoder, built on NVIDIA Parakeet TDT 0.6B v3 and Moondream Parakeet Ultra. It writes punctuation and capitals, and an optional clean mode drops fillers and repeated words. FluidVoice publishes ONNX, MLX, Core ML, and GGUF builds for NVIDIA GPUs, CPUs, and Apple Silicon, and says one GPU runs the model at 3,690 times real time, about an hour of audio in a second. The Apple Silicon package in the Mac app is about 622 MiB. Altic, which publishes the app, reports more than 422,000 downloads and 13,000 GitHub stars. Audio stays on the device unless the user turns on an optional cloud provider for text cleanup. The launch video is a live dictation pass over short spoken questions, including names and places.
Tags
Questions
Frequently asked questions
The FluidVoice launch post reached 44K views on X, with 751 likes, 45 reposts, 28 comments, 797 bookmarks.
FluidVoice launched on X on November 14, 2026 with the message "Sub-1B speech-to-text model for on-device dictation, claiming better accuracy than rivals".
FluidVoice has released Fluid Speech Pro, a 0.6 billion parameter speech-to-text model for on-device dictation on a laptop or phone.
LaunchVideo.com rated the FluidVoice launch 2.00 out of 5: hook 1/5, editing / creativity 1/5, copy 4/5, sentiment of launch 3/5, distribution strategy 1/5.







