# Soumith Chintala says Inkling 2 Small is built for customizability

> Original title: we spent a ton of time understanding what makes models customizable, and built Inkling 2 Small to have the right properties.

- Source: Soumith Chintala
- Published: 2026-11-06T00:24:33.000Z
- Bad Influence: https://badinfluence.news/items/j5uj64chjzkw9744xdgzurza1
- Original: https://x.com/soumithchintala/status/2118499742246465563

## Summary

Soumith Chintala says Inkling 2 Small was built to be customizable, a property the team studied before training. He says that for a fixed data and dollar budget, the model delivers relatively strong results compared with other options. He adds that customizability is a launch-blocking eval, and he invites readers to try the model through Tinker.

## Post

we spent a ton of time understanding what makes models customizable, and built Inkling 2 Small to have the right properties.

For a fixed data and $ budget, this model will get you very good results (relatively) versus other options.

Probably one of the only model programs in the industry with a launch blocking eval around customizability.

Go @tinkerapi !

## Quoted post @lu__jasper

> This the most interesting part of the release to me.
> 
> Inkling-Small was pretty difficult to post-train out the box compared to Chinese models like Qwen. 
> 
> 1/ Good to know it wasn't a skill issue on my part
> 2/ Looks like they specifically tried to address this issue in Inkling 2 training!
> 
> I wonder what tricks went into training the model to be more "customizable".

https://x.com/lu__jasper/status/2118491409879589047