Harmonic, the Robinhood CEO’s AI math startup, launches an AI chatbot app

Harmonic, an AI-startup, co-founded by Robinhood CEO Vlad Teev, announced the beta launch of an iOS and Android-Chatbot app on Monday where users have access to his AI model, Aristotle.

With this launch, the company strives to broaden access to Aristotle, which offer harmonic claims “hallucination-free” answers for questions about mathematical reasoning-a daring claim given the reliability problems of today’s AI models. Harmonic is aimed at creating “mathematical super intelligence” or MSI; The startup ultimately wants to help users with all areas that depend on math, including physics, statistics and computer science.

‘[Aristotle] Is the first product that is available for people that does the reasoning and formally verifies the output, “said Harmonic CEO and co-founder Tudor Achim in an interview with WAN.” Within the domains that Aristotle supports, which are quantitative reasoning domains, we actually guarantee that there are no hallucinations. “

In the end Harmonic says that it is also planning to release an API to give Enterprises access to Aristotle, as well as a web app for consumers.

Harmonic says that Aristotle’s gold medal performance has achieved on the 2025 International Math Olympiade (IMO) via a formal test (which means that the problems were translated into a machine -readable size). Google and OpenAi have also developed AI models that achieved gold medal performance on this year’s IMO, but through informal tests taken in the natural language.

Harmonic says that it does not release other benchmarks for Aristotle at the moment.

The beta launch of Aristotle has only been collected a few weeks since Harmonic $ 100 million in a series B -round has collected under the leadership of Kleiner Perkins with a rating of $ 875 million. Achim claims that Harmonic “continues very quickly” his way to reaching MSI and that investors believed that this was a fair appreciation given the scope of the ambition of his startup.

Various leading technology companies are aimed at training their AI models to solve math problems. Ai who can do maths is in itself valuable, but math is also considered a unique verifiable domain that requires core reasoning skills. Systems that develop these possibilities can also be useful in other domains.

Achim says that Harmonic achieved its hyper -accurate solutions by having Aristotle’s reactions produced in the Open Source Programming language Lean. Before Aristotle gives an answer to users, he says that the model is double checking that the solution is correct via an algorithmic process that does not include AI. The CEO of Harmonic notes that similar technology is used to verify outputs in fields with high deployment such as medical devices and aviation.

Even in a narrow domain, achieving hallucination-free performance of an AI model is an incredibly difficult task. Studies have shown that even leading AI models a lot of hallucinating, and the problem does not seem to be getting better. OpenAI’s newest AI reasoning models hallucinate more than the older ones.