millionways Introduces Thorsten-4, the Foundation Model That Teaches AI to Read the Room and Understand Human Behavior Business Wire Thu, November 12, 2026 at 3:09 PM EST 7 min read META -0.82% AAPL -0.14% Explore stocks on Coinbase Trading disclosure Trading disclosure The above button links to Coinbase. Yahoo Finance is not a broker-dealer or investment adviser and does not offer securities or cryptocurrencies for sale or facilitate trading. Coinbase pays us for certain activity generated through this link. Prices displayed are informational.
Built from four years of repeated interviews with 7,000+ people, Thorsten-4 predicts human intent, motive and likely reactions with 90% validated accuracy.
Thorsten-4 reads how people think, decide, and respond from the way they communicate, then puts that understanding to work wherever a business has AI interacting with humans.
millionways is backed by operators from Meta, Pinterest and Apple and NFL Player DK Metcalf.
NEW YORK, November 12, 2026--(BUSINESS WIRE)--The world's biggest AI labs have built models that can reason, write and speak, but none of them really understand people. That's why millionways, a behavioral intelligence frontier lab, today introduces Thorsten-4, a foundational Large Psychology Model that predicts how humans think, decide and act. Built on 15+ years of proprietary human research, Thorsten-4 gives AI the human understanding it has been missing.
The Intelligence That AI is Missing
AI agents now run sales calls, candidate interviews, client service and negotiations, but the LLMs behind them learned from text. They predict what should be said without modeling why people do what they do, so they miss the hesitation, pushback and motive that decide how a conversation ends. Ten leading LLMs fall 45% short of humans at tracking how a person's thoughts and feelings change over time, and they flip their answers 46% of the time when a user pushes back. Each misread costs a deal, a hire or a client's trust. The same blind spot runs through every product built on these models, from the note taker on a Zoom call to the CRM platform drafting a reply.
Thorsten-4 brings behavioral intelligence to any conversation, whether human-to-human or human-to-AI.
Data No Amount of Compute Can Recreate
Thorsten-4 grew out of 15 years of human research that began long before LLMs existed. Co-founder and CEO Martin Cordsmeier, a bestselling author on human behavior, spent years talking with people from every walk of life, curious about what drives them. He turned that curiosity into a decade-long research program through a nonprofit foundation he built in Germany, where 50+ trained psychologists conducted roughly 7,000 structured interviews with people ranging from an eight-year-old to billionaires. The interviews got participants to open up about their dreams, wishes, and most personal experiences, the kind of honest material that public internet text, much of it AI-written or posted anonymously, doesn't capture. Trained psychologists then hand-tagged 35M+ words from those conversations using established, published methods from psychological research, and the team measured the model's predictions against what participants actually did. Cordsmeier then moved to New York and built millionways around that data.
"Every AI company trains on the same internet, and the internet doesn't record why people do what they do. You can't scrape that, and more compute won't shortcut years of sitting across from real people," said Martin Cordsmeier, co-founder and CEO of millionways. "For the next decade, humans will increasingly communicate with AI in conversations that matter. We're making sure that AI knows who it's talking to."
Thorsten-4 turns how people communicate into a clear read on who they are and what they'll do next. The model works with the conversations companies already have, including meetings, interviews, emails, chats, support calls and earnings calls. It measures each person across 50 behavioral and psychological traits, grounded in decades of motivation and personality research, to show what drives them, how they make decisions, how they handle emotion, and how they react under stress. Thorsten-4's predictions of motivation, decision-making and cognitive preferences were validated against established psychological assessments. Checked against those tests and against what participants actually did over several years, Thorsten-4 reached 90% predictive accuracy without relying on self-reported data alone. Every read Thorsten-4 returns is explainable, tracing back to the behavioral evidence behind it, and it stays consistent when a user pushes back. LLMs work differently. They generate each word based on probability, using patterns learned from massive amounts of public internet text, so even the people who build them can't fully explain a given answer. Ask an LLM the same question twice, or tell it it's wrong, and the answer often changes.
Companies can add Thorsten-4 to the models, agents and AI products they already run, through the millionways Platform, an API, MCP, or a self-hosted deployment.
"Language models learned to predict words. Thorsten-4 learned to predict people and read the room. Understanding people is a different problem from generating language, and it needed its own model built on real behavior," said Max Weidemann, co-founder and CTO of millionways. "Every AI system that deals with people will need a model like this, the same way every system that deals with language needed an LLM."
Applications for Wherever AI Meets People
Thorsten-4 is in use with 10+ companies across executive search, customer experience, financial services and enterprise agentic platforms. The same behavioral intelligence works across industries because the underlying problem is human understanding.
Finance: A multi-strategy hedge fund tested Thorsten-4 against sentiment tools and frontier LLMs, and in Q2 2026 began trading a dedicated $50M portfolio on the model. In testing, a portfolio built on Thorsten-4's reads of earnings calls returned 37% in 2026 through mid-September, before trading costs, and those returns held up after accounting for standard market risk factors.
Customer experience: In completed enterprise pilots, GPT paired with Thorsten-4 outperformed GPT alone at reading customer intent, preferences and decision drivers.
Executive search: Multiple top-10 search firms confirmed Thorsten-4's read on leadership interviews against their own assessments. millionways is building a firm-specific assessment tool with one of them.
Hiring integrity: millionways is developing an Interview Integrity & Authenticity Risk API with an AI recruiting company to flag AI-assisted, coached and proxy answers in AI-led interviews.
"Choosing a leader means betting a company's future on how that person will think, decide and act. millionways brings a deeper understanding of what drives people, how they make decisions and how they respond under pressure. Turning real conversations into explainable behavioral intelligence is a major advance for an industry built on understanding people," said Matt Comyns, co-founder and president of Artico Search.
millionways raised $6 million in U.S. financing and is backed by a mix of tech operators, investors and pro athletes, including Ty Ahmad-Taylor (former VP of Product Marketing, Meta), Dan Gardner (Co-Founder and Executive Chairman, Code and Theory), and Pittsburgh Steelers wide receiver DK Metcalf, among others.
"We backed millionways because we believe behavioral intelligence will become a core capability of AI. Thorsten brings years of original psychological research into real-world decisions, with evidence to back up its predictions. For a board choosing its next CEO, that means understanding how a candidate handles pressure, weighs tough decisions, and leads through uncertainty. As AI agents take on more responsibility across industries, the ability to understand human behavior will only become more valuable. We see millionways building a foundational layer for that future," said Scott Scanlon, CEO and Co-founder, Hunt Scanlon Ventures
Enterprise and AI teams can see Thorsten-4 read real conversations by requesting a demo at millionways.ai.
millionways is a behavioral intelligence frontier lab. Its Large Psychology Model, Thorsten-4, is a foundational model that reads the person behind the words, turning conversations into explainable signals on motive, decision style and likely behavior. The model was built from more than 15 years of psychology research, including four years of repeated interviews with 7,000+ people and 35M+ words tagged by psychologists. Because the underlying problem is human understanding, the same model works across industries, from hiring and wealth management to customer experience and AI agents. Co-founded by Martin Cordsmeier and Max Weidemann, millionways is headquartered in New York and backed by senior operators from Meta, Pinterest and Apple. Learn more at millionways.ai.
View source version on businesswire.com: https://www.businesswire.com/news/home/20261112461856/en/
Media ContactVSC for millionwaysmillionways@vsc.co
