Warsaw – November 09, 2026 -- Ingenix's "Thalamus Reasoning" model ranks first among 29 models on LongevityBench, the independent benchmark for language models in cellular aging research, with a mean rank of 1.42 across the benchmark's 14 ranked tasks. The next best model scores 4.42. The best frontier model, Claude Opus 4.6, scores 9.85.
Ingenix's 9-billion-parameter model wins 11 of 14 benchmark tasks
The model was built by a small team and outperforms 28 published models, including systems from leading frontier labs. It also beats the best non-LLM baseline on eight tasks.
On DNA methylation age prediction, Thalamus Reasoning posts an error of 3.36 years. The best text-prompted model on the same Qwen3.5-9B base scores 12.9 years, and the Horvath clock scores 4.67 years. On transcriptomics ordering, the model reaches 0.828 accuracy against 0.671 for the best other LLM.
Ingenix states that no benchmark data was used in training, which rules out data contamination.
Modality Fusion encodes each biological profile in four tokens
Thalamus Reasoning is part of the company's Biological Reasoning Engine. It converts methylation, expression and proteomics data from complex text into mathematical vectors, the format models use to read images. Ingenix calls the process Modality Fusion. Each profile requires four tokens, while a standard LLM such as ChatGPT needs about 1.3 tokens per word.
CEO Piotr Surma said the ranking "validates our approach of Modality Fusion" and argued that reasoning models deliver meaningful results when they read biological data as measured rather than as text written about it.
Ingenix releases open-source Thalamus Decision on November 9
Ingenix is open-sourcing a version of the model called "Thalamus Decision" on 09 November. Academic and commercial researchers can inspect, reproduce and build on it. The company is also developing its own 95-billion-parameter model.
Ingenix argues that a step change in AI for drug discovery is unlikely to come from hyperscalers tackling every problem at once. It expects the gain from focused teams that combine AI and biology expertise. The company was founded in 2023 and is headquartered in Warsaw. Its leadership comes from the founding team of Applica, the AI company Snowflake acquired in 2022.