The London-based startup beat every human forecaster in a major competition and now wants to sell that edge to hedge funds and governments
A London-based AI startup that can apparently predict the future better than humans just secured $25 million to prove it wasn’t a fluke.
Mantic, founded in 2024, closed a seed round led by Radical Ventures with participation from Microsoft’s M12 venture fund, Balderton Capital, and several angel investors. The money will go toward scaling the company’s AI forecasting models, which specialize in the kind of messy, real-world predictions that spreadsheets and traditional quantitative methods tend to bungle.
Beating humans at their own game
Mantic’s headline credential is its performance in the summer 2026 Metaculus Cup, an online forecasting tournament that pits participants against each other on questions about geopolitics, economics, culture, and policy. The startup’s AI system outperformed every single human entrant.
The only competitor that finished ahead of Mantic was another AI called “laertes.” So the top of the leaderboard was entirely non-human.
What makes Mantic’s approach distinctive is the problem set it targets. The company focuses on what it calls “judgmental forecasting,” predictions that span timeframes from one week to one year, in domains where pure data-driven models historically struggle. Think geopolitical crises, shifting economic policy, or corporate strategy moves where context and nuance matter as much as numbers.
Who’s buying predictions
Mantic’s early customers tell you a lot about where the company sees its market. Hedge funds focused on geopolitical and macroeconomic events have been among the first adopters.
Beyond finance, the company is targeting Fortune 500 companies and government agencies. The pitch to corporates is straightforward: better scenario planning for supply chains, market entry decisions, and risk management. For government clients, the use case centers on anticipating geopolitical developments and policy outcomes.
The $25 million seed round follows a pre-seed raise of approximately $4 million led by Episode 1. DRW, the Chicago-based trading firm, was among the angel backers in that earlier round.
Radical Ventures partner Aaron Rosenberg has joined Mantic’s board following the investment.
The team behind the models
Mantic was co-founded by Toby Shevlane, a former Google DeepMind employee, and Ben Day.
The company’s models are built on frontier AI architectures. The secret sauce appears to be in how those models are fine-tuned, prompted, and calibrated specifically for forecasting tasks, producing probability estimates that are well-calibrated rather than just confident.
