LET’S GOOO 🔥 Just released Smol2Operator, a full open-source recipe...
Amir Mahla@amir_mahla
21 views
Jan 23, 2026
~1 min read
Advertisement
1
LET’S GOOO 🔥 Just released Smol2Operator, a full open-source recipe for turning a 2.2B model into an agentic GUI coder, and all the tools you need to build your own 🫡
0:15
2
We chose a model with ZERO UI understanding. It couldn't even click a button, and we turned it into a fully agentic GUI coder! 🤯
We release the blog that explains all the process hf.co/blog/smol2oper… 📚
As well the final model, datasets, tools and a demo! huggingface.co/collections/sm…
We release the blog that explains all the process hf.co/blog/smol2oper… 📚
As well the final model, datasets, tools and a demo! huggingface.co/collections/sm…
3
The secret sauce? We own the entire process, from preprocessing data to instilling UI understanding and reasoning. End-to-end open-source pipeline
✅ Training recipes
✅ Data processing pipelines
✅ Transformed datasets
✅ The final model
✅ Training recipes
✅ Data processing pipelines
✅ Transformed datasets
✅ The final model
4
Key findings from Smol2Operator👇
1️⃣ Your GUI model’s actions mirror your data’s action space. Design it carefully.
2️⃣ Coordinate choice matters: pixel vs. normalized + image size = big performance shifts.
3️⃣ Reasoning data boosts GUI-grounding accuracy.
1️⃣ Your GUI model’s actions mirror your data’s action space. Design it carefully.
2️⃣ Coordinate choice matters: pixel vs. normalized + image size = big performance shifts.
3️⃣ Reasoning data boosts GUI-grounding accuracy.