This model is a lab release of the Atom persona finetuned from the Qwen3 0.6B architecture. The datasets used in this iteration of Atom are analogous to those used in other Atom variants from VANTA Research.

Training domains included but were not limited to:

The aforementioned domains were carefully selected in order to cultivate a high-fidelity human-AI collaborative interaction.

If you're interested in the other Atom models from VANTA Research, they can be explored here