Vesper models

Run Vesper on your own computer

Small Vesper models you can put inside your game or app. They run offline on the player's own computer, with no API and no cost per message. Or skip the setup and use the hosted API.

Check my computer

Models

Built for games: characters that talk back, fast.

Duet 1Coming soon

A role-play model for NPCs. Give it a character sheet and it stays in character, in short replies.

~110M parameters4k-token memoryNo internet needed
$0.50 one-time
Collab 1Coming soon

Duet's smaller sibling, made for very fast short lines: barks, greetings, quick reactions.

~50M parametersVery fastNo internet needed
$0.50 one-time

Duet and Collab are still being trained. Prices and sizes may change before release.

How fast would it run for you?

A short speed test in your browser (about 2 seconds). Nothing is sent anywhere.

Download or API?

Download
  • Works offline
  • No internet delay
  • No cost per message, ever
  • Speed depends on the player's computer
Hosted API
  • Nothing to install
  • Same speed on every device, even phones
  • $0.02 per million tokens for Duet, $0.01 for Collab
  • Needs an internet connection

Get an API key →

Questions

When are Duet and Collab coming out?

They're next after Sonata 1.5, released together as a "game pack". Follow along on the About page.

What will I get when I download?

The model file plus a small program to run it, with examples for Python and game engines. The license terms come with the release.

Will it run on an old laptop?

Most likely yes. These models are tiny next to big chatbots. Use the check above to see what to expect on your computer.

Why is this check only an estimate?

A browser can only see part of your computer, and a quick test isn't a real game. It's close enough to tell you whether to download or use the API.