Tempus AI Inc. (NASDAQ: TEM) announced initial results from its multimodal foundation model development efforts at the 2026 American Society of Clinical Oncology Annual Meeting. The company reported training its transformer-based model on 2.5 million longitudinal records containing more than 250 million pages of clinical notes, 450,000 digitized medical images, and 500,000 genomic and transcriptomic sequences.

The model demonstrated predictive capabilities in analyzing EGFR-mutant non-small cell lung cancer patients treated with osimertinib. In testing without additional training, the model achieved a C-index of 0.802 for overall survival prediction with a p-value of less than 0.001. The system also showed a hazard ratio of 4.536 between high-risk and low-risk patient subgroups.

Tempus reported that its model produced prognostic value independent of molecular and clinical subgroups, stratifying overall survival in TP53-positive patients with a hazard ratio of 5.96 and progression-free survival in patients without central nervous system metastasis with a hazard ratio of 1.94.

The company stated its models successfully predicted outcomes of patient cohorts that mirrored established clinical trials including KEYNOTE-189, FLAURA-2, and DESTINY in non-small cell lung cancer cases. According to the company, the multimodal patient trajectory model outperformed standard Cox-PH modeling approaches.

"Foundation models, combined with agentic workflows, will help unlock the full potential of precision medicine," said Eric Lefkofsky, founder and CEO of Tempus. The company built its models using more than 500 petabytes of data encompassing over 45 million de-identified patient journeys, with 1.5 million containing sequenced data.

The findings are based on analysis of more than 1.2 million de-identified records with multimodal data, according to the company's press release statement.