Mistral AI for macOS

About

Mistral AI connects TypeWhisper to Mistral’s cloud API, for speech-to-text and for text processing with the same API key. Recordings, files and dictations go to Voxtral Transcribe 3. While you dictate, Voxtral Mini Transcribe Realtime 3 shows live text; when you stop, the recording also goes to Voxtral Transcribe 3, which writes the final text. In our tests it heard dictations more accurately than the realtime model and was done sooner after the stop. Mistral’s text models handle prompts and workflows.

On Windows, the add-on is called Mistral in the app.

Features

• Voxtral Mini Transcribe 3 and Voxtral Small Transcribe 3 for recordings, files and dictations
• Live text while you dictate through Voxtral Mini Transcribe Realtime 3, with the final text from Voxtral Transcribe 3
• 58 languages, with automatic language detection or a fixed language
• Active TypeWhisper dictionary terms sent to Mistral as context, also in live sessions, so names such as “Saoirse Ronan” come out right
• Speaker labels for recordings with several people (Mac)
• Model lists loaded from your Mistral account, so new models appear without an add-on update
• Mistral text models for prompts and workflows, with a reasoning level per model
• Mistral’s own explanation when a request is rejected
• API key stored in the macOS Keychain or, on Windows, encrypted for your Windows account

Transcription Models

Model	ID	Use
Voxtral Mini Transcribe 3	voxtral-mini-transcribe-3-0	Default: recordings, files and the final text of live dictations
Voxtral Small Transcribe 3	voxtral-small-transcribe-3-0	Mistral’s most accurate transcription model, according to Mistral
Voxtral Mini Transcribe Realtime 3	voxtral-mini-transcribe-realtime-3-0	Live text while you dictate

All three accept 58 languages. Saved Voxtral 2 selections such as voxtral-mini-latest switch to the matching Transcribe 3 model. Realtime 2 accepted neither a language nor dictionary terms and is no longer offered.

Text Models

Model	ID	Reasoning levels
Mistral Large 4	mistral-large-4	Low (default), Medium, High
Mistral Medium 3.5	mistral-medium-2604	Off (default), High
Mistral Small 4	mistral-small-2603	Off (default), High
Ministral 3 14B	ministral-14b-2512	–
Ministral 3 8B	ministral-8b-2512	–
Ministral 3 3B	ministral-3b-2512	–

Mistral Small 4 is the default text model. Higher reasoning levels think before they answer, which takes several seconds longer and uses more tokens; the default is the fastest level each model accepts. On Mac, workflows can pick their own level; on Windows, the level set in the add-on also applies to workflows. Saved older models such as mistral-large-latest or pixtral-12b-2409 switch to their current replacement. On Mac, Mistral Medium 3.5 appears once the add-on has loaded your account’s model list.

Prices

Mistral’s list prices in November 2026, per hour of audio:

	Price
Voxtral Mini Transcribe 3	$0.21
Voxtral Small Transcribe 3	$0.24
Voxtral Mini Transcribe Realtime 3	$0.48

Dictations with live text are billed for both the realtime model and the final transcription, about $0.69 per hour with Voxtral Mini Transcribe 3. Text models are billed per token.

Configuration

• API Key - Create an API key in the Mistral console and paste it into the add-on settings. On Mac, saving it also loads your account’s model list.
• Transcription model - Voxtral Mini Transcribe 3 is the default; Voxtral Small Transcribe 3 is more accurate and costs a little more.
• Live text - On Mac, choose Live (default) or After recording under Transcription. On Windows, choose Voxtral Mini Transcribe Realtime 3 as the transcription model; live text needs TypeWhisper 1.3 or newer, and older versions transcribe after you stop.
• Speaker labels (Mac) - Marks who is speaking in recordings with several people. Dictations then transcribe after the recording instead of streaming.
• Text model and reasoning - Choose the model for prompts and workflows, and its reasoning level.
• Temperature - Mistral’s default or a custom value from 0 to 2.
• Refresh (Windows: Refresh models) - Loads the models of your Mistral account.

A model in the list is not a promise that your workspace may use it: Mistral lets a workspace limit a model to zero requests per minute, and requests to it then fail with a rate-limit error.

Setup

1. In TypeWhisper, open Settings > Discover plugins and install Mistral AI. The add-on then appears in the Settings sidebar, and its page opens. TypeWhisper for Windows lists it as Mistral.
2. Enter your Mistral API key and save it.
3. Choose the transcription model, and on Mac whether dictations stream live.
4. Select Mistral AI (on Windows: Mistral) as your transcription engine or as the LLM provider of a workflow.