WARSAW, November 17, 2026 — Polish artificial intelligence startup Ingenix has published performance benchmarks for a biological AI system called Thalamus Decision, which the company says is designed to interpret gene-expression data and support decision-making in biomedical research. Unlike a conventional language model that produces a written response, the system is designed to assign confidence scores to possible outcomes, helping researchers identify when an automated recommendation is sufficiently reliable and when human review is needed. The company’s November 17 announcement, distributed through PR Newswire, outlines the model and its reported benchmarks.

Thalamus Decision is designed to process bulk and single-cell RNA sequencing data. RNA sequencing helps scientists measure gene activity in cells and tissues, providing clues about biological processes, disease states and responses to treatments. Ingenix says the model can support tasks such as identifying cell types, assessing whether samples show signs of malignancy, determining tissue of origin and predicting how cancer cell lines may respond to drugs. HPCwire’s report summarises the model’s stated capabilities.

The company reports that its 9-billion- and 27-billion-parameter versions achieved scores between 84% and 89% on two drug-response prediction tasks using GDSC cell-line data. It also reports approximately 94% performance on gene-essentiality tasks and about 95% accuracy when identifying the type of cancer model being analysed. These figures are company-reported benchmark results and should be interpreted in the context of the datasets, test design and evaluation methods used. Ingenix’s benchmark announcement provides the reported results.

A notable feature is the model’s use of calibrated confidence scores. In biomedical research, a prediction is not useful simply because a system produces an answer; scientists also need to understand how confident the system is and when the evidence is insufficient. Ingenix says its approach could allow a proportion of early-stage drug-development decisions to be automated while referring uncertain cases to human experts. Such a workflow could help researchers prioritise experiments, but it does not replace laboratory validation, clinical trials or regulatory review. HPCwire reports the company’s confidence-calibration claims.

The potential application is important because drug discovery generates large volumes of biological data and requires researchers to decide which targets, compounds and experimental candidates deserve further investigation. AI systems that help interpret gene-expression profiles could support faster screening and reduce some repetitive analysis. Yet biological systems are complex, and performance on selected benchmark datasets does not guarantee that a model will generalise to every disease, patient population or laboratory setting. Independent evaluation will be essential before the technology is relied upon for consequential biomedical decisions. Read the full announcement and its stated limitations.

Ingenix said the 9-billion-parameter version was expected to be open-sourced on November 18, one day after the benchmark announcement. If released as planned, it could give researchers an opportunity to examine the system and test its performance on additional datasets. The wider development illustrates how AI research is moving beyond general-purpose chatbots towards specialised tools built for genomics, molecular biology and translational medicine. HPCwire’s coverage includes the company’s stated release plan.