Dictivo local models and languages

Reference facts about the seven local speech models in the Dictivo desktop app: their sizes, who can use them, how Dictivo picks one for your machine, and the languages you can dictate in.

Last updated November 10, 2026

Applies to: Dictivo 0.3.57 on macOS (Apple Silicon and Intel) and Windows x64. Read from the app's source and the published app by the Dictivo team on 10 November 2026.

Short answer

Dictivo offers seven local speech models, from Tiny at about 75 MB to Large v3 at about 3.1 GB. Tiny is free forever. The other six need the 14-day trial or a Dictivo Local license. During setup, Dictivo reads your memory, processor and GPU and recommends a model for your machine. You can dictate in 11 languages, or let Dictivo detect the language.

What your computer needs · Which Mac suits the larger models

The seven models

Each model is one file that runs on your computer. A larger model needs more disk space and more time for each recording.

The local models in Dictivo, in the order the app lists them
Model	Download size	Who can use it	Dictivo's default for
Tiny	~75 MB	Free forever	Fast on other machines
Base	~142 MB	14-day trial or Local license	Fast on 8-core machines; Medium on other machines
Small	~469 MB	14-day trial or Local license	Fast on GPU machines; Medium on 8-core machines; Quality on other machines
Medium Q5	~540 MB	14-day trial or Local license	Not a default; available to choose
Large v3 Turbo Q5	~600 MB	14-day trial or Local license	Medium on GPU machines; Quality on 8-core machines
Large v3 Turbo	~1.6 GB	14-day trial or Local license	Not a default; available to choose
Large v3	~3.1 GB	14-day trial or Local license	Quality on GPU machines

GPU machines have 16 GB of memory or more and a GPU with 8 GB of memory or more. 8-core machines have 16 GB of memory or more and at least 8 processor cores. Other machines are everything else, including every machine with less than 16 GB of memory.

Without the trial or a Local license, Dictivo uses Tiny for the Fast and Medium tiers.

How Dictivo picks a model

Dictivo offers three speed tiers: Fast, Medium and Quality. The app marks Medium as Recommended. Each tier uses one model.

Dictivo puts your machine in one of three groups by its memory, processor cores and GPU memory. The group sets the model for each tier, as the table above shows.

On Apple Silicon, the GPU shares the Mac's memory. Dictivo counts half the memory as GPU memory, and at least 8 GB. A Mac with 16 GB of memory or more is a GPU machine. A Mac with 8 GB is not.

With the trial or a Local license, Dictivo recommends the Medium model for your group. Without them, it recommends Tiny.

During setup, Dictivo downloads the model you pick, by default the recommended one, and times a short sample with it. It tries the CPU and, where it can use one, the GPU, and keeps the faster. From that time it estimates the other tiers, and marks a tier that would run slowly as "may be slow".

To measure again, for example after a hardware change, click Re-run setup in Settings › Engine.

Memory, processor and GPU in detail

Where models come from and where they live

Dictivo downloads each model from downloads.dictivo.app. If that download fails, it tries huggingface.co. A model download asks only for the model file. It sends no license key and no device id.

Before a download, Dictivo checks the free disk space. It needs the size of the model plus 256 MB.

Dictivo keeps the models in this folder:

Mac: ~/Library/Application Support/Dictivo/private-fast/models
Windows: %LOCALAPPDATA%\Dictivo\private-fast\models

Dictivo also reads models from an older folder, .dictivo/private-fast in your home folder, if an earlier version created it.

On a computer that never goes online, import the model from a file instead. The next section says how.

Every connection Dictivo makes

Languages

You can dictate in 11 languages. Settings › Engine › Speech language lists them with these labels:

English
Deutsch
Français
Español
Italiano
Nederlands
Português
中文
日本語
한국어
Tiếng Việt

The default is Auto-detect: Dictivo detects the language you speak. Choose a fixed language if Dictivo keeps guessing wrong, for example with short sentences or when you mix languages.

With a fixed language, Dictivo also gives the words in your Dictionary to the local model as hints, so it catches names and terms.

The app's own text comes in 10 interface languages: the same list without Tiếng Việt.

Switching and deleting models

Change models in Settings › Engine, under Local model setup. Click a tier to use its model. If the model is not on your computer yet, Dictivo asks before it downloads it.

Advanced — full model catalog, in the same place, lists all seven models. There you can download, select or delete a model, or import one from a file. To import a model:

Open Settings › Engine and then Advanced — full model catalog.
Choose the model in the list next to Import.
Enter the path to the model's .bin file.
Click Import.

Dictivo checks the file's checksum and accepts only the official file for that model. Every model except Tiny needs the trial or a Local license, for a download and for an import.

Delete removes the model file from this computer. You can download or import it again later.

What the models do not do

Dictivo is for live dictation and does not import recorded files.

Dictivo has no mobile app.

The local models work only in Local mode. Cloud Fast, the optional cloud mode, sends a recording to cloud transcription providers instead.

Where your dictation audio goes

All Dictivo docs

Questions

Which model is free?

Tiny, about 75 MB. It stays free forever. The other six models work during the 14-day trial and, after it, with a Dictivo Local license.

Which model should I download first?

The one Dictivo recommends during setup. With the trial or a Local license, that is Large v3 Turbo Q5 on a GPU machine, Small on an 8-core machine and Base on other machines. Without them, it is Tiny.

How much disk space do the models need?

From about 75 MB for Tiny to about 3.1 GB for Large v3, for each model you keep. A download also needs 256 MB of free space on top of the model's size. Delete a model you no longer use to free its space.

Which languages can I dictate in?

English, German, French, Spanish, Italian, Dutch, Portuguese, Chinese, Japanese, Korean and Vietnamese, or Auto-detect. The app's interface comes in the same languages except Vietnamese.

Can I use a model file I already have?

Yes, if it is the official file for one of the seven models. Open Settings › Engine, then Advanced — full model catalog, choose the model, enter the path to its .bin file and click Import. Dictivo checks the checksum and rejects any other file. Every model except Tiny needs the trial or a Local license.