Polish AI company Ingenix tops cellular senescence benchmark, open-sources model

On November 9, Polish AI and biotech company Ingenix announced that its "Thalamus Reasoning" model ranks first among 29 models on the independent benchmark LongevityBench for language models in cellular aging research; the company has also released an open-source version named "Thalamus Decision" for academic and commercial researchers to inspect, reproduce, and continue developing.
According to the company, the model converts methylation, gene expression, and proteomics data into mathematical vectors that can be read directly by the model (a process termed Modality Fusion), with each data profile requiring only 4 tokens, whereas standard large models require approximately 1.3 tokens per word. Benchmark results: average rank of 1.42 across 14 ranking tasks (second-best model 4.42, best-performing frontier model Claude Opus 4.6 scores 9.85), and ranked first in 11 tasks. DNA methylation age error stands at 3.36 years, compared to 12.9 years for the best text-prompt model on the same Qwen3.5-9B base, which also outperforms the Horvath clock at 4.67. Transcriptome sorting accuracy is 0.828, surpassing the 0.671 achieved by other top models. The company emphasized that the training process did not utilize benchmark data, eliminating any risk of data contamination.
In terms of scale, it is a 9-billion parameter model developed by a small team, and the company is currently working on a 95-billion parameter version. CEO Piotr Surma stated that outperforming frontier labs and specialized models on independent benchmarks validates the company's approach: having reasoning models read biological data in the raw format in which it was measured, rather than relying on textual descriptions. Founded in Warsaw in 2023, Ingenix's team originates from Applica, an AI company acquired by Snowflake in 2022, and the firm previously secured a €13 million funding round led by Sofinnova Partners.