Jindu Bio Releases GeneLLM: A Multi-Omics Foundation Model for AI-Driven Science

新智元 · wechat · 2026-08-04

Founded by four Oxford PhDs, Jindu Bio has introduced GeneLLM, a life science foundation model published in Nature Communications. Unlike traditional models relying on annotations, GeneLLM is pretrained directly on raw multi-omics data (RNA, proteins, etc.) using a next-base prediction approach, with its XLarge version reaching 30 billion parameters.

Beyond the model, the company developed BioFord Harness, an infrastructure layer designed to bridge AI with physical laboratories. The system utilizes five collaborative agents to automate the entire research workflow—from literature review and experiment design to equipment scheduling and data analysis. This turns the lab into a programmable system where both successful and failed experiments become valuable training data.

Commercially, Jindu Bio adopts a lightweight physical AI strategy, focusing on orchestrating scientific workflows rather than building heavy assets or competing solely on model size. The startup has closed four funding rounds within a year, including a Series A led by HighLight Capital.

Original post →

More from Venture

Venture channel →