Ingenix Opens Thalamus Decision; Reasoning Beat 28 [2026] Ingenix, a Warsaw-based AI company, released an open-source model called Thalamus Decision on 9 November 2026 for academic and commercial researchers to inspect, reproduce and build on its cellular aging research, according to a press release carried by outlets including WFXR. The release is framed around a reported benchmark result for a sister system, Thalamus Reasoning, a 9 billion parameter model in the Ingenix Biological Reasoning Engine that Ingenix says ranked first among 29 models on LongevityBench, an independent benchmark for language models in cellular aging research; independent verification of that result has not been established in the coverage reviewed. Ingenix also says a larger 95 billion parameter model is under development, with no date, price or license terms disclosed in the sources reviewed. Warsaw-based Ingenix released an open-source AI model called Thalamus Decision on 9 November 2026, according to Inquisitr https://www.inquisitr.com/ingenix-releases-thalamus-decision/ . The release targets academic and commercial researchers who want to inspect, reproduce and build on the company’s work in cellular aging research. It lands alongside a benchmark claim that has drawn most of the early attention. That claim concerns a sister system called Thalamus Reasoning, a 9 billion parameter model that is part of the Ingenix Biological Reasoning Engine. Reports say it ranked first among 29 models on LongevityBench, an independent benchmark for language models in cellular aging research. In plain terms, it finished ahead of 28 others. Independent verification of that result has not been established in the coverage we reviewed. The story matters beyond biology. Open releases are getting narrower and more specialized, and a small European lab is betting that a domain model can earn trust faster than a general chatbot. We saw a similar pattern when LG opened its K-EXAONE 2.5 API for free https://shattered.io/lg-ai-research-k-exaone-2-5-api-free-70-percent-nvidia-2026/ . This analysis separates what Ingenix has confirmed from what remains open, and explains what it means for researchers, developers and investors who follow open-source AI. Don't miss new tech stories on Google Add Shattered once in the Google app and our stories show up in your news suggestions. Add now https://www.google.com/preferences/source?hl=en&q=https%3A%2F%2Fshattered.io%2F What Ingenix Announced on Nov. 9, 2026 The core announcement is short. Ingenix, a company based in Warsaw, Poland, put out an open-source model named Thalamus Decision on 9 November 2026. The press release, carried by outlets such as WFXR https://www.wfxrtv.com/business/press-releases/cision/20261109LN91312/polish-ai-start-up-claims-first-place-on-longevitybench-and-open-sources-a-version-of-its-thalamus-reasoning-model , is titled to say the start-up claims first place on LongevityBench and open-sources a version of its Thalamus Reasoning model. That headline frames the open release as a derivative of the benchmarked system. Ingenix says the goal is transparency. The stated purpose is to let researchers inspect, reproduce and build on the work. The intended audience covers both academic and commercial teams, which matters because it signals that Ingenix wants industry labs to touch the model, not only university groups. The company also says a larger model with 95 billion parameters is under development. No date, price or license terms appear in the sources we reviewed. The table below separates the confirmed items from the open questions, and we use the same split throughout this piece. | Item | Status | Detail from the sources | |---|---|---| | Company and location | Confirmed | Ingenix, Warsaw, Poland | | Open-source model | Confirmed | Thalamus Decision, released 9 November 2026 | | Benchmarked model | Confirmed | Thalamus Reasoning, 9 billion parameters | | Benchmark result | Reported | First among 29 models on LongevityBench | | Larger model | Confirmed as in development | 95 billion parameters, no date given | | Price | Not confirmed | No figure in available sources | | License | Not confirmed | No terms in available sources | | Independent verification | Not established | No outside replication found in coverage | Thalamus Reasoning and the Ingenix Biological Reasoning Engine Thalamus Reasoning is the model that earned the headline. It has 9 billion parameters, a modest size by 2026 standards, and it forms one part of the Ingenix Biological Reasoning Engine. The naming suggests a system built from several pieces, with the language model acting as the reasoning layer inside a larger biology-focused stack. A 9 billion parameter model is small enough to run on a single well-equipped workstation once quantized, though Ingenix has not published hardware requirements for either release. That gap is worth flagging. Researchers choose models partly on what they can afford to run, and a model that needs a cluster serves a different audience than one that fits on a lab desktop. The relationship between Thalamus Reasoning and Thalamus Decision deserves careful reading. The press release title calls the open model a version of Thalamus Reasoning. Beyond that, we have no confirmed detail on how the two differ in size, training data or capability. Anyone assuming the open model matches the benchmarked one is making an assumption the announcement does not support. LongevityBench: What a First Place Among 29 Models Means LongevityBench is described as an independent benchmark for language models working on cellular aging research. Thalamus Reasoning ranked first among 29 models, ahead of 28 others, according to reports including a Techflow Post https://www.techflowpost.com/en-US/newsletter/143679 item on the Polish company topping a cellular senescence benchmark. What the ranking tells you A first-place finish in a field of 29 is a meaningful signal if the field was fair. It says the model handled the benchmark’s tasks better than the other entrants did, at least under the conditions Ingenix describes. For a 9 billion parameter system, beating larger and better-funded rivals would count as a real result. The ranking does not tell you how the model behaves on your own data. Benchmarks measure a fixed set of tasks, and cellular aging research is a wide field. A strong score on one suite predicts performance on adjacent work only loosely. Why independent verification still matters The benchmark is called independent, but the claim that Thalamus Reasoning won comes from the company’s own announcement. Nothing we found shows a third party rerunning the evaluation. That is normal for a release day, and it is also the standard reason to wait before treating any result as settled. Teams that want to test the claim themselves can borrow a method from our walkthrough on building a local LLM-as-a-judge evaluation harness https://shattered.io/llm-as-a-judge-local-eval-harness-12-steps-2026/ . Run the same tasks, log the prompts, and compare against a baseline you trust. Open Source Without a License: The Missing Details The phrase open source carries weight, and the sources we reviewed do not confirm the license. That omission is the biggest practical unknown for commercial users. Permissive terms allow a company to build a product on the weights. Restrictive terms can block exactly the commercial use Ingenix says it welcomes. Several other details are also unconfirmed. The architecture is not described in the coverage we saw. Training data is not described. Hosting location and download availability are unclear, and so are the terms for commercial usage. Until Ingenix or a repository page spells these out, readers should treat the release as announced rather than fully documented. This matters more in biology than in many fields. Training data for a model that reasons about cellular aging may include human-derived datasets with their own consent and usage conditions. A clean license on the weights does not automatically clear the data that shaped them, so legal teams at commercial labs will want a full account before they build on it. Who Can Use Thalamus Decision: Academic and Commercial Researchers Ingenix names two groups, academic and commercial researchers. The pairing is deliberate. Academic labs bring reproducibility and peer scrutiny, while commercial teams bring budgets and applied problems. By inviting both, Ingenix positions Thalamus Decision as shared infrastructure rather than a closed product. For a university lab, the appeal is practical. An open model can be inspected layer by layer, rerun on new data, and cited in a paper with a precise version. For a biotech startup, it offers a head start without paying per-token fees to a hosted service, assuming the license allows it. Developers who plan to host the model should budget for memory early. Our guide to tracking per-process VRAM and LLM cache use https://shattered.io/gputaskman-per-process-vram-llm-kv-cache-mit-v0-3-3-2026/ shows how quickly context length can eat a GPU, a problem that applies to any model in this size class. The 95 Billion Parameter Model Ingenix Is Building Ingenix says it is developing an additional model with 95 billion parameters. That is more than ten times the size of Thalamus Reasoning. If the company carries its benchmark approach to that scale, the gap between its open and closed offerings could widen. The coverage, including a write-up on AI News Bites https://ainewsbites.com/ingenix-thalamus-longevitybench-open-source/ , also covers the open-source release and the company’s next model. No release date, pricing or licensing plan has been confirmed. Whether the 95 billion parameter version will follow the open-source path of Thalamus Decision is also unstated. That uncertainty shapes strategy for anyone adopting the smaller model today. If Ingenix keeps its largest system proprietary, the open release works as a showcase and a funnel. If it opens the big model too, the release becomes a genuine platform. Researchers should avoid building workflows that assume either outcome. Why a 9 Billion Parameter Model Can Matter in Biology Bigger is not always better in a narrow field. A model tuned for cellular aging tasks can beat a far larger general system if its training and design fit the problem. The reported LongevityBench result, if it holds, would support that view. It would show that a 9 billion parameter model can outrank 28 others when the test is specific enough. Size also shapes who can participate. A 9 billion parameter system sits in the range that small labs can often fine-tune and serve without a data center contract. That lowers the barrier for a postdoc team in a university or a three-person biotech start-up. A 95 billion parameter system would demand far more memory and budget, which is one reason the two models likely serve different customers. There is a flip side. Narrow models can fail in surprising ways outside their home turf. A system that excels on ranked tasks about cellular aging may stumble on a neighboring question about drug interactions or clinical language. Researchers should map their own questions against the benchmark’s scope before assuming the ranking transfers. Open Weights Versus a Closed API: The Trade-Offs Ingenix chose to release a model that teams can inspect, which differs from offering only a hosted endpoint. The trade-offs cut both ways, and they are worth laying out for anyone deciding how to adopt it. Open weights give a lab control. It can freeze a version, so results stay stable across a long study. It can keep sensitive data on its own hardware, a real advantage when working with human-derived datasets. And it can modify the model, which a closed API rarely allows. Those benefits explain why the stated purpose of the release centers on inspection and reproduction. A closed API has its own appeal. It removes the burden of hardware, patching and scaling, and it lets a vendor update the model without user effort. The cost is dependence. A lab that builds on someone else’s endpoint inherits its pricing, its outages and its policy changes. In regulated research settings, that dependence can become a compliance headache. The best choice depends on the project. Short exploratory work may suit a hosted model. Long reproducible studies lean toward open weights. Since Ingenix has confirmed neither the license nor the hosting details for Thalamus Decision, teams should hold off on committing a multi-year study to it until those points are settled. How Thalamus Decision Compares With Other Open Releases Direct comparison is hard because open models in 2026 target very different jobs and use different yardsticks. Ingenix competes on a biology benchmark, while others compete on coding or general language scores. The table lists recent open or open-access releases we covered, using only the claims reported at the time. | Model | Maker | Focus | Headline claim as reported | |---|---|---|---| | Thalamus Decision | Ingenix | Cellular aging research | Sister model Thalamus Reasoning ranked first of 29 on LongevityBench | | K-EXAONE 2.5 | LG AI Research | General language, API | API opened free, 70% gain claimed | | Riemann 13.1B | Not named in our report | General language model | NLL fell from 1.323 to 1.310 via staged growth | | Ripple-65B | Not named in our report | Coding agents | 80.5 on Terminal-Bench 2.1 | The numbers in the last column cannot be compared with each other. A 70% gain, a loss figure and a ranking measure different things on different tests. What the table does show is a market splitting into specialties. Our coverage of the Riemann 13.1B staged-growth model https://shattered.io/riemann-technical-account-13-1b-staged-growth-nll-1-310-2026/ and of the Ripple-65B coding result https://shattered.io/ripple-65b-preview-terminal-bench-2-1-80-5-open-coding-2026/ illustrates the same trend from the general and coding sides. Ingenix competes on trust within a niche rather than on breadth. For readers tracking the wider field, our ranking of the best open-source LLMs from October 2026 https://shattered.io/best-open-source-llm-october-2026-kimi-glm-deepseek/ covers the general-purpose side, and the AI and machine learning section https://shattered.io/ai-machine-learning/ collects the rest. Market Impact: Who Gains and Who Watches Ingenix is a start-up, and we found no report of a listed company reacting to the news. The impact is quieter. It shows up in how research teams budget for tools, and in how investors judge small labs that publish weights. The likely winners are research groups that lacked access to a purpose-built biology model. They gain a free option to test. The likely watchers are larger AI vendors with general models and enterprise biology offerings. They now face a specialist that claims a top ranking on a domain test, backed by an open release that invites direct comparison. For Ingenix itself, the open release carries a business logic. Open weights spread a name quickly among researchers, who then become the buyers or collaborators for the company’s other products. The risk is that a good open model lets customers skip a paid tier. The 95 billion parameter model is where that tension will be resolved, since it may be the premium product. A hub like Hakky Handbook https://book.st-hakky.com/en/news/polish-ai-startup-tops-longevity-benchmark , which logged the release as a Polish start-up topping a longevity benchmark, shows how fast niche AI news now travels between aggregators. Small labs gain visibility fast, and they also face scrutiny fast. Historical Context: Why Domain Benchmarks Draw Skepticism Model makers have long picked benchmarks that flatter them, and readers have learned to ask who ran the test. Domain-specific suites raise the stakes because fewer people can judge them. A general benchmark gets thousands of eyes. A cellular aging benchmark may get a few dozen. That history explains the careful wording around the Ingenix result. Calling LongevityBench independent is a good start, but the field has seen cases where a benchmark’s tasks leaked into training data or where a leaderboard rewarded format tricks over real skill. Without a published training data description, outsiders cannot rule out overlap. The healthy response is neither cheering nor dismissal. Open weights give the community a way to check, which is the main reason the release is more interesting than a pure benchmark announcement. A claim that can be tested on a public model is stronger than one that sits behind an API. A Checklist for Teams Evaluating the Release Teams considering Thalamus Decision can run a short due-diligence pass. The table maps each check to its reason and to what the available sources say today. | Check | Why it matters | Status in sources | |---|---|---| | Read the license text | Decides whether commercial use is allowed | Not confirmed | | Confirm where weights are hosted | Affects trust and download integrity | Not confirmed | | Check hardware requirements | Sets your serving cost | Not confirmed | | Review training data notes | Flags consent and leakage risks | Not confirmed | | Rerun LongevityBench tasks | Tests the ranking claim directly | No outside replication found | | Compare with your current baseline | Shows value on your own data | Your own test | Keep a log of every test run so results can be reproduced later. A minimal record might look like the snippet below. The values in angle brackets are placeholders for you to fill in once you have the actual repository and version details. { "model": "Thalamus Decision", "source": "