Tempus AI, Inc. (NASDAQ: TEM) announced the publication of study results in Nature Medicine demonstrating that its PRISM2 Pathology Foundation Model performs clinically important diagnostic tasks with high precision. Developed in collaboration with Microsoft, the multimodal slide-level model outperformed or matched existing foundation models across a comprehensive set of diagnostic, biomarker, and patient outcome prediction tasks. This development is significant for precision oncology as it enables the prediction of long-term outcomes, including colorectal cancer recurrence-free survival, without requiring specialized fine-tuning for complex research tasks.

The findings highlight the technical capabilities of PRISM2 as part of Tempus’ proprietary Pathology Foundation Models. The model transforms routine hematoxylin and eosin (H&E) slides into deep biological insights by combining large vision models built from pathology images with large language models. This multimodal approach unlocks diagnostic-grade precision in research involving cancer detection, biomarker identification, and prognosis prediction. Razik Yousfi, Senior Vice President and General Manager of AI Products at Tempus, stated that PRISM2 represents a leap forward in scale and multi-modal AI capabilities.

PRISM2’s performance is underpinned by its extensive training data. The model was trained on a diverse set of 2.3 million whole-slide images and 14 million diagnostic question-answer pairs derived from nearly 700,000 pathology reports. This dataset is described as the largest multimodal slide-level pathology dataset to date. By aligning whole-slide pathology images with the language of clinical diagnosis through clinical dialogue training, the model can seamlessly handle complex diagnostic and prognostic research tasks.

Key Training Data Metrics

According to management, the Pathology Foundation Models allow for a detailed understanding of tissue, unlocking clinical-grade precision and novel research applications. These applications are capable of predicting patient outcomes and biomarker status. The ability to predict colorectal cancer recurrence-free survival with high performance when further tuned specifically for long-term outcomes distinguishes PRISM2 from standalone models.

To support ongoing research, open science, and other non-commercial and non-clinical use cases, the full PRISM2 model weights are publicly available through Hugging Face. This accessibility aims to facilitate broader adoption and validation within the scientific community.

What the Numbers Show

The scale of the training data directly correlates with the model's reported versatility. With 14 million diagnostic question-answer pairs derived from nearly 700,000 reports, the dataset density suggests a robust alignment between visual pathology features and clinical language. This large-scale multimodal training allows PRISM2 to perform across multiple task types—diagnostic, biomarker, and prognostic—without specialized fine-tuning, indicating a generalizable architecture rather than a narrow, task-specific tool.

Tempus AI (NASDAQ: TEM) reported second-quarter revenue of $382.5 million, rising 22% year-over-year and surpassing Street consensus estimates of $379.7 million. Despite the top-line beat and a narrower-than-expected loss per share of four cents versus an estimated 14 cents loss, shares declined 2.4% in after-hours trading to $43.25. The market reaction suggests investors are pricing in concerns over decelerating growth momentum relative to recent quarters and the modest nature of the raised full-year guidance.

The company’s performance was driven by strong demand across its core segments. Revenue from the Diagnostics segment reached $289.3 million, up 20% year-over-year, while Data and Applications revenue grew 28% year-over-year to $93.2 million. Oncology volume growth accelerated by 31% year-over-year, and Data Licensing & Modeling revenue surged 36% year-over-year. Tempus CEO Eric Lefkofsky attributed the results to investments in AI driving growth in the company’s two largest businesses: Oncology Diagnostics and Data Licensing.

Financial Performance Breakdown

Beyond organic growth, Tempus secured approximately $200 million in new data and applications licenses during the quarter. This contract inflow supports the trajectory of its high-margin data business, which continues to expand faster than the diagnostics segment.

Forward Guidance and Market Reaction

Tempus raised its full-year revenue guidance from a prior range of $1.59 billion to $1.60 billion to a new range of $1.595 billion to $1.605 billion. This adjustment implies approximately 25% year-over-year growth for the full year. Adjusted EBITDA is guided at around $65 million. Notably, this guidance excludes any impact from the pending acquisition of Personalis, which is expected to close in late fourth quarter of 2026 or early 2027.

Technical Outlook and Key Levels

From a longer-term trend perspective, TEM is in a downtrend, trading 15.1% below the 20-day SMA, 14.9% below the 50-day SMA, 13.5% below the 100-day SMA, and 26.9% below the 200-day SMA. Momentum indicators show MACD below its signal line with a negative histogram, pointing to fading upside pressure. The 50-day SMA remains below the 200-day SMA, a death cross that occurred in January, keeping intermediate-term trend traders cautious. With the stock down 23.40% over the past 12 months, key resistance sits at $51.00 near the 50-day SMA area, while support is located at $42.50.

What the Numbers Show

The divergence between the earnings beat and the stock price decline highlights investor sensitivity to growth rates rather than absolute profitability metrics. While the company beat estimates, the 22% revenue growth represents a moderation compared to prior periods, leading the market to interpret the minimal upward revision in guidance as a signal of potential slowdown in the latter half of the year. The acquisition of Personalis remains a key variable for future scale, but its exclusion from current guidance leaves near-term growth dependent on existing operations.