Polish AI company Ingenix tops cellular aging benchmark and open-sources a model version

TechFlow reported on November 9 that Polish AI and biotechnology company Ingenix announced its “Thalamus Reasoning” model ranked first among 29 models on LongevityBench, an independent benchmark for language models in cellular aging research; the company also released an open-source version named “Thalamus Decision” for academic and commercial researchers to inspect, reproduce, and further develop. According to the company, the model converts methylation, gene expression, and proteomics data into mathematical vectors directly readable by the model (a process the company calls Modality Fusion), requiring only 4 tokens per data profile, whereas a standard large model needs about 1.3 tokens per word. Benchmark results: across 14 ranked tasks, average rank 1.42 (next-best model 4.42, best-performing frontier model Claude Opus 4.6 scored 9.85), and it ranked first in 11 tasks; DNA methylation age error was 3.36 years, compared with 12.9 years for the best text-prompted model on the same Qwen3.5-9B base and 4.67 for the Horvath clock; transcriptome ranking accuracy was 0.828, versus 0.671 for other best models. The company emphasized that benchmark data were not used during training, so there was no data contamination. In terms of scale, this is a 9 billion-parameter model built by a small team, and the company is developing a 95 billion-parameter version. CEO Piotr Surma said leading frontier labs and specialized models on an independent benchmark validated the company’s approach—having reasoning models read biological data in the raw way it is measured, rather than reading text about it. Ingenix was founded in Warsaw in 2023; its team comes from the founding team of AI company Applica, acquired by Snowflake in 2022, and previously secured a EUR 13 million funding round led by Sofinnova Partners.

Source: TechFlow