What is Kosmos, new AI research tool Sam Altman calls ‘most important impact of AI’?

“The core innovation in Kosmos is our use of structured world models, which allow us to efficiently incorporate information extracted over hundreds of agent trajectories and maintain coherence towards a specific research objective over tens of millions of tokens,” the company said in a blog post published last week.

Calling it an exciting innovation, Altman wrote in a post on X, “I expect we are going to see a lot more things like this and it will be one of the most important impacts of AI. Congrats to the Future House team.”

Notably, the AI research tool was able to reproduce unpublished scientific work on a single run, while it took human scientists roughly four months to complete the same research. The buzz around Kosmos comes at a time when several AI startups are racing to develop AI research tools specifically for the scientific domain.

This is exciting; I expect we are going to see a lot more things like this and it will be one of the most important impacts of AI. Congrats to the Future House team.https://t.co/Cxeh8UlWdk

— Sam Altman (@sama) November 16, 2025

Earlier this year, Google unveiled an “AI co-scientist”, designed to aid scientists in creating hypotheses and experimental research plans. More importantly, Sam Altman and his counterparts in the tech industry have argued that such AI tools will be able to accelerate scientific discovery in fields like medicine at a much faster pace.
Despite the warning of an AI market bubble, Altman has previously said that he believes the technology will have a positive, long-term impact in fields like mathematics and science.

On the other hand, several researchers have questioned whether today’s AI tools are ready for such scientific work, as large language models (LLMs) that power most of these tools are still prone to errors and hallucinations. Academics can currently use Kosmos for free, but with certain limitations on usage. Its subscription plan is priced at 200 credits per run, equivalent to $200.

Multiple runs might be needed for certain projects as Kosmos can produce dead ends or follow statistically interesting but scientifically irrelevant patterns, according to Edison Scientific.

The company said the tool has been designed with transparency and traceability in mind. This means that “every conclusion in a Kosmos report can be traced through our platform to the specific lines of code or the specific passages in the scientific literature that inspired it, ensuring that Kosmos’ reports are fully auditable at all times”.

“We have also validated Kosmos by using it both to replicate unpublished findings and to make net new contributions to the scientific literature,” it added.

How has Kosmos performed in testing?

As part of its testing, Edison Scientific said it used Kosmos to publish scientific discoveries in various fields such as neuroscience, materials science, genetics, and ageing research.

It was able to generate findings that matched three scientific findings previously made by researchers, including unpublished work on nucleotide metabolism changes in the brains of hypothermic mice. Kosmos was also used to propose the following four novel scientific insights:

– Using higher levels of the antioxidant enzyme SOD2 to reduce heart fibrosis in humans.
– A new molecular explanation for how a genetic variant may lower the risk of Type 2 diabetes.
– A new method for analysing how tau proteins accumulate in the brain in Alzheimer’s disease.
– A large-scale analysis of ageing in neurons found that certain neurons in the entorhinal cortex lose expression of flippase genes with age.

“The most surprising part of our work on Kosmos—for us, at least—was our finding that a single Kosmos run can accomplish work equivalent to 6 months of a PhD or postdoctoral scientist,” Edison Scientific said. However, the company also said that several of these findings are still being validated by conducting wet lab experiments.