LeCun's New Paper: 109M-Parameter Model Beats 72B LLM in Representation

Yann LeCun and Randall Balestriero, in their new Paris lab after LeCun's departure from Meta, published a paper showing that a 109M-parameter model outperforms a 72B-parameter LLM in world-understanding representation tasks.

2026-11-02 ~ 2026-11-02 · 2 related posts

• LeCun's new paper claims a 109M-parameter model beats a 72B model — heyshrutimishra · 2026-11-02
• LeCun's new paper: 109M-param text JEPA beats 72B LLMs at aligning with vision — heyshrutimishra · 2026-11-02