Skip to content
ZenVoice
Download

Legacy models

Retired Models
Still Yours

Kanso Koi 1.0, Kanso Satori 1.0 and Kanso Koi 0.9 have left the Models page. Install and choose them from Terminal.

The Legacy Models

Kept exactly as they were released, and still measured.

Status
Retired from the app
Install with
zenvoice formatting
Language
English
Replaced by
Kanso Sumi

  • Kanso Koi 1.0Legacy

    The newer Koi, on a 4-bit Qwen3-1.7B base with a live adapter.

    kanso-koi-1.0

    0.97% WER · 0.35 s · 1.05 GB

  • Kanso Satori 1.0Legacy

    The 4B Kanso model, on Qwen3-4B-Instruct-2507.

    kanso-satori-1.0

    0.43% WER · 0.47 s · 2.27 GB

  • Kanso Koi 0.9Legacy

    The original Koi, on a 6-bit Qwen3-1.7B base.

    kanso-koi-0.9

    2.06% WER · 0.46 s · 1.55 GB

WER and median time on Kanso Dictation Bench V1, the same run as the Benchmarks page →. Already chose one before it was retired? Nothing changes for you.

Set One Up

Four steps in Terminal and ZenVoice. Swap in kanso-koi-1.0 or kanso-koi-0.9 for Koi.

  1. 1

    Install the command

    In ZenVoice, open Settings → Transcribe and install the command-line tool.

  2. 2

    Download the model

    From its pinned source; every file's size and SHA-256 is checked.

    $ zenvoice formatting install kanso-satori-1.0
  3. 3

    Choose it

    Dictation and recordings use it from the next dictation.

    $ zenvoice formatting use kanso-satori-1.0
  4. 4

    Turn on Polished

    In ZenVoice → Formatting. Clean and Verbatim don't use a formatting model.

Commands

One command manages every formatting model, current and legacy. Add --json to any of them for scripts.

  • zenvoice formatting

    Lists every formatting model, marks the one in use with *, and labels the legacy ones.

  • zenvoice formatting install <id>

    Downloads a model and checks it. Run it again to resume an interrupted download.

  • zenvoice formatting use <id>

    Makes it the model Polished formatting uses.

  • zenvoice formatting use kanso-sumi-1.0

    Goes back to a current model. Choosing any model in the app does the same.

  • zenvoice formatting remove <id>

    Deletes a model that isn't in use, to free the disk space.

  • zenvoice doctor

    Shows which formatting model is in use and whether it's installed.

Installing and choosing a model need your ZenVoice licence, as transcribing from Terminal does. A model in use can’t be removed; choose another first.

In the App

ZenVoice shows a legacy model only while it is the one you use.

Shown while chosen

While a legacy model is your choice, the Models page shows it, marked Legacy, and the Formatting page names it.

Gone when you switch

Choose any other formatting model in ZenVoice and the legacy one leaves the app. Only zenvoice formatting use brings it back.

Never a download

The app never offers a legacy model. If its files are missing, dictation uses Apple Intelligence, or the rules, until you install it again.

See every model →Benchmarks