360Brew is the 150-billion-parameter decoder-only foundation model LinkedIn deployed in 2026 to replace its multi-source feed ranking with a unified, semantic, interest-graph-based content recommendation system.
Built on 📝LLaMA 3 architecture and publicly confirmed in March 2026, 360Brew represents every member and every piece of content as a vector in a shared semantic space, then matches them by inferred topical interest rather than first-degree network proximity. A companion Generative Recommender transformer reads each user's last ~1,000 interactions as a chronological sequence, allowing personalization to work from day one for new members with no engagement history.
The shift collapsed most of the 2025 LinkedIn growth playbook. Posts now compete inside topical interest clusters rather than network graphs, dwell time and substantive long-form comments dominate ranking signals, hashtags are deprecated as a primary distribution mechanism, and topic consistency across a creator's post history determines amplification. Document-format carousels lead all formats while native video reach has collapsed as the model down-ranks low-dwell content.
For creators and marketers, 360Brew rewards narrow topical authority over broad network reach. The implications for content strategy under the new feed are mapped in detail in 📝The 7 Things You Need to Know About LinkedIn's Algorithm in 2026.
The shift from network logic to interest logic forced a rewrite of my entire LinkedIn Post Creator prompt — every assumption baked into the 2025 version was either neutered or actively penalized under 360Brew. The most useful reframe: stop optimizing for who reads it, start optimizing for whether anyone stays on the page for a full minute.