Ingenix Claims Top LongevityBench Ranking, Opens Thalamus Model
Polish AI start-up Ingenix says its 9-billion-parameter Thalamus Reasoning model ranks first among 29 systems on LongevityBench, a benchmark for language models in cellular aging research. The company is also open-sourcing Thalamus Decision, a version intended for academic and commercial researchers.
- Thalamus Reasoning ranks first among 29 models on LongevityBench, according to Ingenix.
- The model ranked first on 11 of the benchmark's 14 tasks.
- Ingenix is open-sourcing a version called Thalamus Decision on 09 November.
- The company is developing a 95-billion-parameter model.
- Ingenix says no LongevityBench data was used to train Thalamus Reasoning.
Thalamus leads LongevityBench results
According to Ingenix, Thalamus Reasoning placed first across LongevityBench's 14 ranked tasks, with a mean rank of 1.42. The next best model had a mean rank of 4.42, while Claude Opus 4.6, identified by the company as the best frontier model, scored 9.85.
Ingenix says Thalamus Reasoning ranked first on 11 of the 14 tasks and finished ahead of the best non-LLM baseline on eight. The company also reports a DNA methylation age error of 3.36 years, compared with 12.9 years for the best text-prompted model using the same Qwen3.5-9B base. The result was also better than the Horvath clock's 4.67 years.
On transcriptomics ordering, the model achieved an accuracy of 0.828, compared with 0.671 for the best other LLM. Ingenix says no benchmark data was used to train Thalamus Reasoning, so the evaluation did not involve data contamination.
A biological data approach based on Modality Fusion
Thalamus Reasoning is part of Ingenix's Biological Reasoning Engine. The model transforms methylation, expression and proteomics data from complex text into mathematical vectors, which the company describes as the native language models use to read images.
Ingenix calls this process Modality Fusion. Each profile generated through the process requires four tokens, according to the company. For comparison, Ingenix says a standard LLM such as ChatGPT requires around 1.3 tokens per word.
The company says its architecture integrates models across modalities and biological scales, then reasons directly across their representations. Its engine is designed for translational and clinical research and development decisions, with outputs that include a reasoning trace covering mechanism, evidence and caveats.
“Ranking ahead of frontier labs and specialised models on an independent benchmark validates our approach of Modality Fusion,” Piotr Surma, CEO of Ingenix, said in the announcement. “It shows that when a reasoning model reads biological data as it was measured, rather than as text written about it, meaningful results can be delivered.”
Open-source release and the company's next model
Ingenix said it is releasing Thalamus Decision on 09 November, giving academic and commercial researchers access to inspect, reproduce and build on the work. The release makes a version of the model available outside the company, although the announcement does not specify a license or hosting platform.
The start-up is also developing a 95-billion-parameter model. Ingenix was founded in 2023 and is headquartered in Warsaw. Its founding team came from Applica, an AI company acquired by Snowflake in 2022. The company says Applica's TILT model contributed to the GPT-4 benchmark.
For researchers in cellular aging and related areas, the release provides an opportunity to examine Ingenix's approach and test its benchmark claims. More broadly, the announcement presents a case for specialised AI systems that combine machine learning expertise with biology knowledge, rather than relying only on general-purpose models. The available evidence in the announcement is based on Ingenix's reported LongevityBench results.
