Acacia Graph Foundation Model Trained on Common Crawl
September 28, 2026
Acacia is a graph foundation model trained from scratch on the web graph, bypassing the need for LLM integration. It supports arbitrary feature dimensions and tasks like node classification and link prediction via in-context learning without additional training or classification heads.
HOW THIS AFFECTS YOU
●
builderYou can use this for zero-shot graph tasks without the latency or cost of stitching in a separate LLM.
●
researcherThis demonstrates that emergent capabilities can be acquired by graph models trained from scratch without LLM assistance.