Limite ships 1B math model trained on under 300B tokens

tensorqt · x · 2026-09-22

Limite has released Limite 1B - Violetto, a tiny model pitched at "high-frequency mathematical intelligence", with just 1B parameters and under 300B training tokens. Researchers like YouJiacheng praised the data efficiency as remarkable, drawing attention to the small-model route.

Related event: Paradigm Releases First AI Model Limite 1B, Targeting High-Frequency Mathematical Intelligence(11 posts)→

Original post →

More from Models

Models channel →