Netflix releases GenRec, an LLM-native recommendation model that achieves statistically significant improvements over traditional models

GenRec is an LLM-backed recommendation ranker that post-trains an internal foundation LLM on Netflix-specific data and objectives.

GenRec is an LLM-backed recommendation ranker that post-trains an internal foundation LLM on Netflix-specific data and objectives. It verbalizes user histories, item metadata, and context as text, and adds a catalog-aware scoring head over Netflix titles. The model is trained with a multi-objective loss that combines a recommendation ranking objective, language modeling objectives, and alignment via reward-weighted training. In a large-scale A/B test, GenRec achieved statistically significant improvements in both short-term and long-term online metrics, while using only a small fraction of the Phase-2 labeled data and input signals. The model's architecture is based on a decoder-only Transformer, and it is served on Netflix's internal LLM stack using vLLM. The release of GenRec is an early but promising step towards a more LLM-centric recommendation stack at Netflix, with potential benefits including improved personalization, increased efficiency, and enhanced user experience. The model's performance is attributed to its ability to capture rich relationships in a shared semantic space, and its alignment with long-term member satisfaction and business goals. As Netflix continues to develop and refine GenRec, it is likely to have a significant impact on the company's recommendation capabilities, and potentially pave the way for further innovation in the field of LLM-native recommendation systems.