---
title: README
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
---

# Vertex Research

**Open weights, no matter what.**

Vertex Research exists for one reason: to make sure open weight models keep getting made. As frontier labs release fewer and fewer open models, we train our own - from scratch, fully open, Apache 2.0 - so that anyone can always download, run, study, and build on a model that nobody can take away.

## Why

The industry is drifting toward closed APIs. Weights that used to be published now stay behind endpoints. If that trend continues, the ability to run a model on your own machine — offline, private, forever — depends on groups that keep releasing weights anyway. We intend to be one of them.

## What we do

- Train compact language models from scratch and release every usable artifact: base, instruct, and reasoning variants, plus GGUF quantizations for llama.cpp.
- Keep everything Apache 2.0. No gated downloads, no acceptance forms, no API-only releases.
- Document what worked and what didn't, so others can train their own.

## Models

| Model | Parameters | Status |
|-------|------------|--------|
| Vertex 0.6 100M | 100M | Released (base 1024–8192 ctx, instruct, GGUF) |
| Vertex 0.6 35M | 35M | Released |
| Vertex 0.1 10M | 10M | Released |

More sizes and versions are on the way. Follow the org to stay updated.

## In training

- Vertex 0.6 100M v2

## Roadmap

(Top model is the next to train)

- [ ] 200M preview
- [ ] 15M (multi-turn chat focus)
- [ ] 10M
- [ ] 50M

---

*Vertex Research: small models, sharp edges, weights you keep.*