Meta’s Generative Ads Model (GEM): Scaling the Central Model and Compute Engine for Frontier Ads AI

Meta's 2026 generation of GEM shifts from per-domain optimizations to a single central foundation model that learns across all Meta ads platforms, delivering an 18.9% conversion improvement and 12% ad CTR improvement across surfaces.

The model uses a new LLM-style architecture called Kunlun, co-designed with the latest GPU hardware, enabling 2.5x model complexity growth and LLM-like scaling laws.

A post-training technique called LoopFM transfers GEM's representations to downstream models at no additional compute cost or latency.