> ## 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.

# Choose a model and create your first mode

> Create your first Superwhisper mode and pick its models along the way. A Message mode works with cloud or local models.

Modes help you customize communication styles across apps and use cases. Every mode includes the voice model that transcribes, and a language model and prompt that shapes the result.

## Create a Message mode

**Message** mode turns your dictation into a ready-to-send message, with a tone slider from casual to formal.

**1. Open Modes**

Open Superwhisper's settings and click **Modes** in the sidebar.

**2. Create the mode**

Open **Modes** and click **Create mode** (the sparkle button at the top right). Under **Preset**, select **Message**. Click the mode's name to rename it.

**3. Pick your models**

Two setups work well:

> [tab] Cloud (start here)
>
> Keep the defaults: **S1-Voice** for transcription and **S1-Language** for formatting. Nothing to download.
>
> ![Superwhisper Message mode using the S1-Voice and S1-Language cloud models](https://mintcdn.com/superwhisper/SFDRA4G70egO6eAv/images/screens/mac/modes-editor-cloud.png?fit=max\&auto=format\&n=SFDRA4G70egO6eAv\&q=85\&s=90a52d2c3c55d9fe33f5235841631831)

> [tab] Local
>
> Set **Voice Model** to **Cohere Transcribe** and **Language Model** to **S1-mini**. Both run on your device: nothing you dictate leaves it. First use downloads the models (Cohere Transcribe is 1.3 GB).
>
> ![Superwhisper Message mode running fully on-device with Cohere Transcribe and S1-mini](https://mintcdn.com/superwhisper/do-XBuvOUgvVmZf6/images/screens/mac/modes-editor-local.png?fit=max\&auto=format\&n=do-XBuvOUgvVmZf6\&q=85\&s=8ec2bd55b0e15ccb29944e72e63ceeaa)

**4. Dictate**

Click into a text field, press **⌥ Space** (Option + Space), and speak. Superwhisper records with your active mode, the one marked with a green dot in the Modes list. Press **⌥ ⇧ K** (Option + Shift + K) to switch modes.

## Explore more models

Open **Settings → Models library** to browse every model, download local ones, and remove ones you no longer use.

![Superwhisper Models library listing voice and language models with download options](https://mintcdn.com/superwhisper/SFDRA4G70egO6eAv/images/screens/mac/models-library-voice.png?fit=max\&auto=format\&n=SFDRA4G70egO6eAv\&q=85\&s=08109f1c8cf062312a278ef00faccec1)

### Cloud vs. local

| | Cloud | Local |
| :- | :- | :- |
| Setup | None; works immediately | One-time download (75 MB – 3 GB) |
| Privacy | Audio processed in the cloud (zero data retention policy) | Audio never leaves your device |
| Offline use | Requires internet | Works offline |
| Speed | Fast on any hardware | Fast on modern hardware |
| Best for | Most users, older machines | Sensitive work, travel, privacy-conscious users |

### Our picks

| Model | Runs | Size | Why pick it |
| :- | :- | :- | :- |
| **S1-Voice** | Cloud | No download | The most accurate model we ship (6.6% word error rate) |
| **Cohere Transcribe** | Local | 1.3 GB | The most accurate local model (8.4% word error rate); our first pick for on-device |
| **Parakeet Multilanguage** | Local | 494 MB | A fast local option on Apple silicon and the only local voice model for Windows ARM. Best for multilingual dictation. |
| **Whisper Large v3 Turbo** | Local | 1.6 GB | Runs well on Windows machines with a Vulkan-capable GPU |

Accuracy numbers come from [our public benchmarks](https://superwhisper.com/benchmarks). For speed comparisons, see [benchmarks](https://superwhisper.com/docs/models/benchmarks.md).

**Every Whisper model is free**, including Whisper Large v3 Turbo. Cohere Transcribe, the Parakeet models, and all cloud models need [Pro](https://superwhisper.com/docs/billing/plans.md).

**On Windows:** Cohere Transcribe and Whisper Large v3 Turbo use Vulkan when your GPU supports it, and fall back to the much slower CPU otherwise. Windows ARM (Snapdragon) devices don't support them yet. On ARM, use Parakeet Multilanguage or a cloud model. Details in [Windows requirements](https://superwhisper.com/docs/get-started/windows.md#requirements).

For every model's languages, speed, accuracy, and size, see [Voice models](https://superwhisper.com/docs/models/voice.md) and [Language models](https://superwhisper.com/docs/models/language.md).

## Next step

> [card] [Understand modes](https://superwhisper.com/docs/modes/modes.md)
>
> Modes decide what happens to your transcription: plain text, a formatted message, an email, or a custom transformation.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.