Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design
Meta Andromeda achieves 100x throughput lift, scoring over one billion ads per second via model-hardware co-design
“surfaces the top candidates at over one billion ads per second, a new throughput benchmark for ads retrieval-stage models (i.e., a 100x lift compared to the 2024 Meta Andromeda system)”
Meta's next-generation Andromeda ads retrieval engine scores over one billion ads per second by co-designing the ML model, system architecture, and GPU hardware together — a 100x throughput improvement over the 2024 system. The key breakthrough is bringing near ranking-level model quality to the retrieval stage, previously too expensive to apply at retrieval scale. This matters as a signal of how frontier ML systems engineering is evolving to handle generative-AI-driven candidate pool explosions, though it is specific to ads infrastructure rather than broadly applicable AI research.