About
What it is for
A far larger and more capable model than Wren, built on Google's open Gemma and fine-tuned by Avenpa for Korean and English chat. In development; not yet available.
Kestrel is our high-performance model. Instead of training from scratch, we start from Google's open Gemma and fine-tune it ourselves for Korean and English conversation.
It is much larger than Wren, so it needs far more compute, and it is still being built. We will publish what it can and cannot do before it opens for chat.
In short
Kestrel, briefly
01
Built on Gemma
Starts from Google's open model instead of training from scratch.
02
Fine-tuned in-house
Avenpa adapts it for Korean and English conversation.
03
In development
Not available yet; limits will be published first.
Questions about Kestrel?
Write to the studio. A person who works on it will answer.
Contact the studio