Harmonic, an AI startup from Robinhood CEO Vlad Tenev, has launched a beta chatbot for its math AI, Aristotle. The company claims 'hallucination-free' results by using a formal verification process, a major step towards reliable AI.
In a world where artificial intelligence models can write poetry and generate stunning images, their persistent struggle with basic logic and mathematics remains a critical flaw. AI “hallucinations” — instances where models confidently state incorrect information — have become a significant barrier to their deployment in high-stakes fields. A new contender, Harmonic, co-founded by Robinhood CEO Vlad Tenev, is stepping into the arena with a bold claim: to have developed a “hallucination-free” AI for mathematical reasoning.
Harmonic, founded in 2023 by Tenev and former Helm.ai CTO Tudor Achim, is not trying to build another general-purpose chatbot. Its mission is far more specific and ambitious: to create what it calls “mathematical superintelligence” (MSI). This focus on the rigorous, verifiable world of mathematics has attracted serious attention and capital. The Palo Alto-based startup recently secured $100 million in a Series B funding round led by Kleiner Perkins, bringing its valuation to a hefty $875 million.
While Tenev is a well-known figure in the finance world, he also holds degrees in mathematics from Stanford and UCLA. He serves as Harmonic's executive chairman, with Achim leading the company as CEO. Tenev has made it clear that their goal extends beyond just acing math tests. “The near-term goal is to build an AI that solves math problems at a level that is superior to any human,” Tenev told Bloomberg News. “The ultimate goal would be to solve major unsolved mathematical problems and expand that to problems in physics and computer science.”
The first product from this venture is Aristotle, an AI model now accessible to the public through a beta chatbot app on iOS and Android. What sets Aristotle apart is its foundational promise of accuracy. Harmonic CEO Tudor Achim made a powerful statement in an interview with TechCrunch: “[Aristotle] is the first product available to people that does reasoning and formally verifies the output. Within the domains that Aristotle supports, which are quantitative reasoning domains, we actually do guarantee that there’s no hallucinations.”
This is a stark contrast to the probabilistic nature of large language models (LLMs) from giants like OpenAI and Google, which have been shown to struggle with reliability, sometimes with newer models hallucinating even more than their predecessors.
Harmonic's confidence stems from its unique technical approach. Instead of just predicting the next word in a sentence, Aristotle is engineered to produce its solutions in Lean, an open-source programming language designed for mathematical proofs. Before an answer is presented to the user, it undergoes a rigorous, non-AI algorithmic check to ensure every step of the reasoning is logically sound. This process, known as formal verification, is already a standard in safety-critical industries like aviation and medical device manufacturing, where failure is not an option.
“We can ensure that every piece of output and every step in our model’s reasoning is verifiably correct,” Tenev explained. “That is just a very different approach to building AI models that I think is going to be the approach that dominates in the future.”
To back up its claims, Harmonic has put Aristotle to the test. The model achieved a gold medal performance on the 2025 International Math Olympiad (IMO). Crucially, this was done through a formal test where problems were translated into a machine-readable format, allowing for the kind of rigorous verification that is central to Harmonic's philosophy. This differs from the informal, natural language tests undertaken by other AI models that have also claimed high scores at the IMO.
— Harmonic (@HarmonicMath) July 28, 2025
For those interested in experiencing this new approach to AI reasoning, Harmonic has made it straightforward to get started. The company has released a beta version of its chatbot app, which provides direct access to the Aristotle model.
The company also plans to release an API for enterprise use and a web application for consumers in the future, which will broaden access to its technology.
Harmonic is making a calculated bet that the future of AI, especially in professional and scientific contexts, hinges on trust and verifiability. By starting with the universal and logical language of mathematics, the company is building a foundation that could potentially be extended to other complex, logic-based fields. While the challenge of scaling this “hallucination-free” approach to the messy, ambiguous nature of human language and real-world problems remains immense, Harmonic’s work represents a significant development. It suggests a future where AI tools are not just creatively proficient but are also demonstrably correct, a shift that could redefine their role across science, engineering, and finance.
As if it weren't enough that humans have struggled with prime numbers for centuries, now we have to drag silicon into it. While the rest of us are trying to understand poetry or sarcasm, Aristotle is focused on whether 2+2 really is 4. Sure, it's important, but it's not exactly a Shakespearean sonnet. It's a bit like building a supercomputer just to check grocery receipts. Precise, but rather... narrow-minded.
But on a more serious note, the hidden potential lies precisely in this "boring" precision. Let's stop thinking about calculators and look further. Imagine the pharmaceutical industry. Protein modeling or simulating drug effects is currently an incredibly expensive and lengthy process, full of dead ends. An AI that can not only design but also mathematically prove the stability of a molecular structure could shorten drug development by years and save billions. Instead of trial and error, we would have mathematical certainty even before the first molecule is synthesized in a lab.
Or consider urban planning and logistics. What about designing an entire city's transportation system that is mathematically optimized for minimal congestion and maximum safety? Aristotle could not only propose a solution but also formally verify that, under given conditions, a total collapse of the network cannot occur. This would change urban planning from the ground up.
And what about the creative industries? Mathematics and music are inextricably linked. Imagine a tool that doesn't compose music based on statistics, like today's generative models do, but on the formal mathematical rules of harmony and counterpoint. It could create perfectly structured, yet novel, works in the style of Bach, where every single note has a logical justification.
So yes, maybe it starts with counting apples and pears. But if Harmonic succeeds, we might soon be verifying whether the universe is just a very complex mathematical equation. And that's a thought worth pondering, isn't it?
