Polish AI Start-up Claims First Place on LongevityBench and Open-Sources a Version of its Thalamus Reasoning Model

• A 9-billion-parameter model built by a small team outperforms 28 published models, including the leading frontier-lab systems.
• Ingenix is now open-sourcing a version of this model, while developing its own 95-billion-parameter model.

WARSAW, Poland, Nov. 9, 2026 /CNW/ -- Ingenix has today announced that its model, "Thalamus Reasoning", which forms part of its Biological Reasoning Engine, now ranks first out of 29 models on LongevityBench, the independent benchmark for language models in cellular aging research. The company also confirmed it will release an open-source version of Thalamus, called "Thalamus Decision", putting the model in the hands of researchers worldwide.

Benchmark results

Thalamus Reasoning transforms methylation, expression and proteomics data from complex text into easier-to-understand mathematical vectors, the native "language" that models use to read images. Each profile it generates through this process, which the company calls Modality Fusion, requires just four tokens (a standard LLM like ChatGPT requires around 1.3 tokens per word), making it highly efficient.

Key performance metrics include:

• First of 29 overall across LongevityBench's 14 ranked tasks, with a mean rank of 1.42. The next best model scores 4.42; the best frontier model, Claude Opus 4.6, scores 9.85. 
• First on 11 of 14 tasks, and ahead of the best non-LLM baseline on eight.
• DNA methylation age error of 3.36 years, against 12.9 years for the best text-prompted model on the same Qwen3.5-9B base, and better than the Horvath clock (4.67). 
• Transcriptomics ordering accuracy of 0.828, against 0.671 for the best other LLM.

Importantly, no benchmark data was used in training Thalamus Reasoning, meaning no data contamination occurred.

Open Source, and a Team that Understands Biology's Complexity

Ingenix is open-sourcing Thalamus Decision on 09 November so that academic and commercial researchers can inspect, reproduce and build on its work.

Together, these results mark a wider point in the ever-progressing TechBio space: a step change in AI for drug discovery is not likely to come from hyperscalers looking to tackle every problem at once, but instead from a more focused team who are directly combining expertise in AI with expertise in biology. Ingenix's founding team were among the first to bring multimodal models to the general market, and are now applying the same method of thinking to biology.

"Ranking ahead of frontier labs and specialised models on an independent benchmark validates our approach of Modality Fusion," said Piotr Surma, CEO of Ingenix. "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. Modality Fusion is what makes that possible, and we're excited to see what researchers do with it."

About Ingenix

Ingenix was founded around a single question: what would it take for AI to truly understand biology? The Ingenix Biological Reasoning Engine is the company's answer.  Built on Modality Fusion, a novel AI architecture that integrates best-in-class models across modalities and biological scales, then reasons across their representations directly. Ingenix is the first to name this approach and the first to bring it to biology. The engine is built for translational and clinical R&D decisions, and every output arrives with the reasoning trace behind it: mechanism, evidence, and the caveats laid out alongside the conclusions. Founded in 2023 and headquartered in Warsaw, Ingenix is led by AI researchers and engineers from the founding team of Applica, the AI company acquired by Snowflake in 2022, whose TILT model contributed to the GPT-4 benchmark. For more information, visit ingenix.ai.

Media Contact

Barnaby Pickering, Director, 59 North Communications
[e-mail]

SOURCE Ingenix