ByteDance's X-Rec does generative retrieval via flow matching with 3.46x throughput

_reachsumit · x · 2026-09-25

ByteDance's X-Rec technical report models recommendation as flow matching in continuous item embedding space, generating multiple retrieval triggers in parallel instead of quantizing items into discrete semantic IDs.

On a streaming benchmark it substantially outperforms U2I baselines, matches SID-AR retrieval quality, and delivers 3.46x higher inference throughput.

Original post →

More from Research

Research channel →