Meta scales GEM ads model to single foundation for all platforms

Meta published engineering details on October 13, 2026, describing the expansion of its Generative Ads Model (GEM) into a unified recommendation system. The company shifted from separate optimization models for individual apps to a single central foundation model that processes data across its entire advertising ecosystem. This architectural change aims to improve ad relevance and performance simultaneously across all user surfaces.

The updated GEM system has generated significant performance metrics since its introduction last year. According to Meta’s engineering blog, the model delivered an 18.9% improvement in conversions and a 12% increase in ad click-through rates. These gains occurred across multiple ad surfaces and objectives, indicating that the centralized approach effectively handles diverse advertising goals without degrading performance on specific platforms.

Technical Architecture and Scaling

Meta addressed challenges in traditional recommender systems by adopting a large language model-inspired paradigm. The company developed an architecture named Kunlun, co-designed with latest-generation GPU hardware to support larger and deeper models. This design allows for long-context learning over user engagement history and achieves a 2.5x increase in model complexity compared to earlier versions of GEM.

The system utilizes a technique called LoopFM for efficient intelligence transfer to downstream models. This post-training method shares GEM’s internal representations and sequential memory with user-facing models. According to the source material, this approach doubles intelligence transfer efficiency without adding compute costs or latency, enabling faster response times for advertisers and users.

Platform Impact and User Experience

The 2026 generation of GEM operates across dozens of surfaces, including Facebook, Instagram, Reels, and Stories. Previously, each surface maintained independent prediction systems, creating challenges for simultaneous performance improvements. The new central compute engine jointly predicts clicks and conversions for all key advertiser objectives, from mobile app installs to off-site conversions.

Specific platform data shows varied uplifts in conversion rates for targeted users. Meta reported a 5.8% conversion increase on Facebook, a 4.1% rise on Instagram, and a 1.6% improvement for off-site conversions. These figures demonstrate that the unified model maintains effectiveness across different funnel stages, from initial awareness to final purchase actions.

What should you do?

Security and technology professionals should monitor Meta’s engineering updates for changes in data collection practices. Organizations using Meta’s advertising platforms should review their tracking pixels and conversion APIs to ensure compatibility with the new central model. Advertisers should analyze their campaign performance metrics to identify anomalies caused by the system-wide algorithm shift.

What to do and how to stay safe: Meta

• Review advertising campaign metrics for significant deviations in click-through or conversion rates following the update.
• Verify that tracking pixels and conversion APIs are functioning correctly with the new unified prediction engine.
• Monitor Meta’s engineering blog for further documentation on data handling and model updates.
• Audit cross-platform ad strategies to align with the new full-funnel optimization capabilities.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is the Generative Ads Model (GEM)?

GEM is Meta’s recommendation system foundation model that uses an LLM-inspired paradigm to optimize ad delivery across its platforms.

How much did conversions improve with the new GEM model?

The model delivered an 18.9% conversion improvement and a 12% ad click-through rate improvement across Meta’s ad surfaces.

What is the Kunlun architecture?

Kunlun is an LLM-style architecture co-designed with GPU hardware that enables 2.5x model complexity growth and efficient scaling.