Twitter (X) Open Sources Its Recommendation Algorithm: Inside the 48M Parameter Model Behind "For You"

No more guessing games about shadowbans or visibility weights. The X Engineering team has just released a detailed technical breakdown (along with the GitHub repo) that fully reveals the black box powering your timeline.

This post offers a hardcore look at how they utilize SimClusters and Real Graph models to distill, score, and rank content from 500 million daily tweets down to the handful you actually see. Whether you're researching ML architectures or just trying to reverse-engineer the engagement logic, this is the definitive primary source.

Source: X Engineering Blog