> ## Documentation Index > Fetch the complete documentation index at: https://superwhisper.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Voice > Compare Superwhisper's voice models, cloud and local. See which languages, speed, accuracy, and size each one offers. Voice models turn your recorded audio into text. There are two types: cloud models run on hosted servers with no download, and local models run entirely on your device. Every mode pairs one voice model with an optional [language model](https://superwhisper.com/docs/models/language.md). **Which languages are supported?** It depends on the voice model. S1-Voice and the multilingual Whisper models cover 100+ languages. Nova and Scribe support the languages shown in your mode's language menu. Parakeet Multilanguage covers 24. Cohere Transcribe covers 14: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese, and Arabic. Parakeet and the Whisper (English) models are English only. Pick a language in your mode's settings to see what the selected model supports. > [!NOTE] > > Speed and accuracy columns are relative ratings from 1 to 10, for comparing models at a glance. Measured word error rates and latency come from our benchmark suite: see [How We Benchmark](https://superwhisper.com/docs/models/benchmarks.md) for the method, or [the leaderboard](https://superwhisper.com/benchmarks) for every number. ## Superwhisper (Cloud) Hosted by Superwhisper and optimized for low latency and accuracy. Works in 100+ languages with nothing to download. | Name | Lang | Translate | Speed | Accuracy | Size | License | | - | - | - | - | - | - | - | | S1-Voice | multi | ✖ | 10 | 9 | Cloud | Pro | S1-Voice has the lowest word error rate on our benchmarks, 6.6% across the eight benchmark datasets. See [How We Benchmark](https://superwhisper.com/docs/models/benchmarks.md) for what that measures. ## Cohere Transcribe (Local) The most accurate local model, with an 8.4% word error rate on Apple M4. It supports 14 languages (listed above), doesn't translate, and doesn't support realtime transcription. It runs through MLX at 4-bit precision on your device's GPU: Apple silicon on macOS, Vulkan on Windows. It isn't available on Windows ARM. | Name | Lang | Translate | Speed | Accuracy | Size | License | | - | - | - | - | - | - | - | | Cohere Transcribe | multi | ✖ | 9 | 10 | 1.3 GB | Pro | ## Nvidia Parakeet (Local) Based on Nvidia's [Parakeet](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2) models, running locally. On macOS they run through Argmax's [WhisperKit SDK](https://github.com/argmaxinc/whisperkit); on Windows they run through ONNX. They process long recordings in parallel. They can struggle with punctuation and occasionally hallucinate on single-word recordings. | Name | Lang | Translate | Speed | Accuracy | Size | License | | - | - | - | - | - | - | - | | Parakeet | en | ✖ | 10 | 8 | 476 MB | Pro | | Parakeet Multilanguage | multi | ✖ | 10 | 8 | 494 MB | Pro | Parakeet Multilanguage covers 24 languages. On iOS, a **Parakeet Japanese** model is also available. > [!NOTE] > > Parakeet Multilanguage is the only local option on Windows ARM (Snapdragon) devices, which don't support Vulkan. See [Windows requirements](https://superwhisper.com/docs/get-started/windows.md#requirements). ## Whisper Models (Local) Based on the [Whisper](https://github.com/openai/whisper) model series from OpenAI, running locally through [whisper.cpp](https://github.com/ggerganov/whisper.cpp). | Name | Lang | Translate | Speed | Accuracy | Size | License | | - | - | - | - | - | - | - | | Whisper Large v3 Turbo | all | ✖ | 8 | 8 | 1.6 GB | Free | | Whisper Large v3 Turbo (Chinese) | zh | ✖ | 8 | 8 | 1.6 GB | Free | | Whisper Large v3 | all | ✔ | 5 | 10 | 3 GB | Free | | Whisper Large | all | ✔ | 5 | 10 | 3 GB | Free | | Whisper Medium | all | ✔ | 6 | 8 | 1.5 GB | Free | | Whisper Medium (English) | en | ✖ | 6 | 8 | 1.5 GB | Free | | Whisper Small | all | ✔ | 8 | 5 | 500 MB | Free | | Whisper Small (English) | en | ✖ | 8 | 5 | 500 MB | Free | | Whisper Base | all | ✔ | 9 | 3 | 150 MB | Free | | Whisper Base (English) | en | ✖ | 9 | 3 | 150 MB | Free | | Whisper Tiny | all | ✔ | 10 | 1 | 75 MB | Free | | Whisper Tiny (English) | en | ✖ | 10 | 1 | 75 MB | Free | **Every Whisper model is free**, including Large v3 Turbo. They're the models to reach for on the [Free plan](https://superwhisper.com/docs/billing/plans.md), and they run entirely on your device. Whisper Large v3 and the Chinese Turbo build are marked experimental in the app. > [!NOTE] > > These models were renamed in the app. Whisper Large was formerly Superwhisper Ultra, Whisper Large v3 Turbo was Ultra V3 Turbo, and the small models were Standard, Nano, and Fast. ## Deepgram (Cloud) The Nova series of cloud models, hosted by Deepgram. Nova models support [realtime transcription](https://superwhisper.com/docs/common-issues/realtime.md) and give the most accurate speaker labels with **Identify Speakers**. Nova 3 is also available in [Meetings](https://superwhisper.com/docs/meetings/overview.md#choose-models). | Name | Lang | Translate | Speed | Accuracy | Size | License | | - | - | - | - | - | - | - | | Nova 3 | multi | ✖ | 7 | 8 | Cloud | Pro | | Nova 2 | multi | ✖ | 7 | 7 | Cloud | Pro | | Nova Medical | en | ✖ | 10 | 7 | Cloud | Pro | ## ElevenLabs (Cloud) Scribe is ElevenLabs Scribe V2, hosted by ElevenLabs. It supports realtime transcription and detects the language automatically. | Name | Lang | Translate | Speed | Accuracy | Size | License | | - | - | - | - | - | - | - | | Scribe | multi | ✖ | 8 | 10 | Cloud | Pro | ## Realtime transcription Realtime shows results as you record. It works with the Nova models, Scribe, and the Parakeet models. On Windows, only the experimental **Parakeet Real-time** models support it, not the standard Parakeet model. See [Realtime on Windows](https://superwhisper.com/docs/common-issues/realtime.md#realtime-on-windows). Cohere Transcribe and the Whisper models don't support it. On Nvidia (Parakeet) and ElevenLabs models, you can't turn on Realtime and **Identify Speakers** in the same mode. ## Choosing a model Not sure where to start? [Create Your First Mode](https://superwhisper.com/docs/get-started/choose-your-model.md#our-picks) has our short list: S1-Voice for cloud, Cohere Transcribe or Parakeet Multilanguage for local. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.