Which local speech-to-text models are people actually using?
By Hugging Face downloads over the last 30 days, Whisper turbo leads the 26 reviewed models with 5,052,286 downloads, followed by Whisper base (4,902,763). This tracker combines those public figures with likes, repository activity and the speech-recognition models currently trending on the Hub.
Snapshot from 11 Nov 2026; current figures load automatically. Popularity is not quality: use it alongside the matcher and your own benchmark.
On this page
Reviewed models by Hugging Face downloads
Click a column heading to sort. Moonshine Tiny Streaming (English) and Moonshine Medium Streaming (English) are distributed outside the Hub, so no figures are shown.
| Model | Downloads (30 days) | All-time downloads | Likes | Last updated | Repository |
|---|---|---|---|---|---|
| Whisper turbo OpenAI · 809M | 5,052,286 | 127,832,256 | 3,520 | 4 Oct 2024 | openai/whisper-large-v3-turbo ↗ |
| Whisper base OpenAI · 74M | 4,902,763 | 55,597,021 | 313 | 29 Feb 2024 | openai/whisper-base ↗ |
| Whisper small OpenAI · 244M | 4,429,963 | 142,207,032 | 647 | 29 Feb 2024 | openai/whisper-small ↗ |
| Qwen3-ASR 1.7B Qwen · 1.7B | 1,667,507 | 18,241,522 | 1,237 | 30 Jan 2026 | Qwen/Qwen3-ASR-1.7B ↗ |
| Voxtral Mini 4B Realtime 2602 Mistral AI · 4B | 1,539,259 | 14,784,796 | 1,062 | 11 Mar 2026 | mistralai/Voxtral-Mini-4B-Realtime-2602 ↗ |
| Distil-Whisper large-v3 Hugging Face / Distil-Whisper · large-v3 distilled | 740,910 | 21,569,535 | 386 | 21 Apr 2026 | distil-whisper/distil-large-v3 ↗ |
| VibeVoice-ASR 7B Microsoft · 9B total | 714,453 | 6,622,297 | 1,328 | 27 Jan 2026 | microsoft/VibeVoice-ASR ↗ |
| Whisper tiny OpenAI · 39M | 632,368 | 23,987,506 | 472 | 29 Feb 2024 | openai/whisper-tiny ↗ |
| Qwen3-ASR 0.6B Qwen · 0.6B | 584,234 | 9,894,790 | 399 | 30 Jan 2026 | Qwen/Qwen3-ASR-0.6B ↗ |
| Nemotron 3.5 ASR Streaming 0.6B NVIDIA · 600M | 539,841 | 4,283,426 | 1,269 | 10 Sept 2026 | nvidia/nemotron-3.5-asr-streaming-0.6b ↗ |
| Parakeet TDT 0.6B v3 NVIDIA · 600M | 529,600 | 3,895,041 | 1,270 | 5 Aug 2026 | nvidia/parakeet-tdt-0.6b-v3 ↗ |
| Whisper medium OpenAI · 769M | 359,082 | 20,830,088 | 317 | 29 Feb 2024 | openai/whisper-medium ↗ |
| Cohere Transcribe 03-2026 Cohere Labs · 2B | 106,517 | 3,556,230 | 1,238 | 7 Oct 2026 | CohereLabs/cohere-transcribe-03-2026 ↗ |
| VibeVoice-ASR-BitNet Microsoft · 0.3B quantized | 99,595 | 206,926 | 214 | 24 Jul 2026 | microsoft/VibeVoice-ASR-BitNet ↗ |
| Whistle Cactus Compute · 16.9 MB file (parameters not stated) | 90,000 | 110,664 | 566 | 5 Nov 2026 | Cactus-Compute/whistle ↗ |
| Granite Speech 5.0 470M TurboCTC IBM · 470M | 55,633 | 144,654 | 116 | 6 Nov 2026 | ibm-granite/granite-speech-5.0-470m-turboctc ↗ |
| Orukeet Oruk AI · 627M | 33,850 | 79,956 | 109 | 24 Sept 2026 | oruk/orukeet ↗ |
| Jais-ASR IFM · 5.7B total | 10,751 | 10,760 | 22 | 13 Oct 2026 | IFM/Jais-ASR ↗ |
| Fun-ASR-MLT-Nano-2512 FunAudioLLM · 800M | 9,896 | 19,764 | 72 | 5 Aug 2026 | FunAudioLLM/Fun-ASR-MLT-Nano-2512 ↗ |
| Granite Speech 5.0 2B IBM · 2B | 9,852 | 9,855 | 82 | 6 Nov 2026 | ibm-granite/granite-speech-5.0-2b ↗ |
| Parakeet Redux Moondream · 1.58-bit Parakeet 0.6B (178 MB) | 9,581 | 26,581 | 317 | 23 Sept 2026 | moondream/parakeet-redux ↗ |
| VibeVoice-ASR-Streaming-7B Microsoft · 9B total | 7,826 | 15,554 | 268 | 3 Sept 2026 | microsoft/VibeVoice-ASR-Streaming-7B ↗ |
| VibeVoice-ASR Streaming 1.5B Microsoft · 3B total | 1,810 | 14,765 | 66 | 3 Sept 2026 | microsoft/VibeVoice-ASR-Streaming-1.5B ↗ |
| Omnilingual ASR CTC 300M Meta AI · 325M | 0 | 0 | 17 | 27 Nov 2025 | facebook/omniASR-CTC-300M ↗ |
Speech-recognition models trending on Hugging Face
The Hub's own trending list for the automatic-speech-recognition task. Models we have not reviewed are marked; trending is not an endorsement, and fine-tunes or demos often appear here.
New releases we have checked are on the new models page.
Runtime activity on GitHub
Repository stars, latest tagged release and last push for the runtimes our profiles document, from the GitHub REST API (11 Nov 2026).
| Runtime | Stars | Latest release | Released | Last push |
|---|---|---|---|---|
| Hugging Face Transformers ↗ huggingface/transformers | 168,811 | v5.22.2 | 6 Nov 2026 | 11 Nov 2026 |
| openai-whisper ↗ openai/whisper | 112,450 | v20250625 | 26 Jun 2025 | 31 Aug 2026 |
| vLLM ↗ vllm-project/vllm | 96,741 | v0.33.0rc3 | 2 Nov 2026 | 11 Nov 2026 |
| VibeVoice ↗ microsoft/VibeVoice | 57,008 | No GitHub release | — | 8 Oct 2026 |
| whisper.cpp ↗ ggml-org/whisper.cpp | 55,092 | v1.9.5 | 6 Oct 2026 | 10 Nov 2026 |
| faster-whisper ↗ SYSTRAN/faster-whisper | 26,664 | v1.2.1 | 31 Oct 2025 | 10 Nov 2026 |
| FunASR ↗ modelscope/FunASR | 20,998 | v1.4.25 | 11 Nov 2026 | 11 Nov 2026 |
| NVIDIA NeMo ↗ NVIDIA/NeMo | 18,734 | v3.0.0 | 7 Aug 2026 | 11 Nov 2026 |
| sherpa-onnx ↗ k2-fsa/sherpa-onnx | 15,883 | v2.0.1 | 26 Oct 2026 | 11 Nov 2026 |
| Moonshine ↗ moonshine-ai/moonshine | 11,473 | v0.1.6 | 6 Nov 2026 | 10 Nov 2026 |
| mlx-audio ↗ Blaizzy/mlx-audio | 8,248 | v0.6.5 | 9 Nov 2026 | 10 Nov 2026 |
| Qwen3-ASR ↗ QwenLM/Qwen3-ASR | 3,877 | No GitHub release | — | 26 Jun 2026 |
| Omnilingual ASR ↗ facebookresearch/omnilingual-asr | 2,967 | 0.2.0 | 22 Dec 2025 | 30 Dec 2025 |
| VibeASR.cpp ↗ microsoft/VibeASR.cpp | 277 | No GitHub release | — | 15 Sept 2026 |
How to read these numbers
- Downloads (30 days) are file downloads counted by Hugging Face over the last 30 days. Automated pipelines, mirrors and CI jobs count too, so they indicate reach, not active users or quality.
- Older, smaller checkpoints such as Whisper tiny and base are embedded in many tools and tests, which inflates their downloads relative to newer models.
- Likes are bookmarks by Hugging Face users. Last updated is the latest change to the model repository, which may be a README edit rather than new weights.
- Trending is Hugging Face's own ranking of recent interest for the task.
- None of these figures change the matcher's scores, which come from verified capabilities and conservative ratings (methodology).
Data from the Hugging Face Hub API ↗ and the GitHub REST API ↗. Figures are cached for a few hours. Model names and repositories belong to their owners.