Mantic raises $25M to build AI that predicts the future better than humans

A London-based AI startup called Mantic just closed a $25M seed round to do something deceptively simple: predict what’s going to happen next. Not stock prices in the next five milliseconds. Not tomorrow’s weather. The messy, medium-term stuff, like geopolitical shifts, economic turning points, and the kinds of uncertain events that keep strategists up at night.

The round was led by Radical Ventures, with Microsoft’s M12 and Balderton Capital also participating.

Beating humans at their own game

The company’s calling card is its performance in the Metaculus Cup this past summer. Mantic’s AI system beat every single human participant. It also outperformed nearly all competing bots.

Mantic calls its approach “judgmental forecasting.” The term comes from the forecasting literature and refers to predictions that rely on human-style reasoning, research, and judgment rather than purely statistical pattern matching. The company’s system combines this reasoning framework with historical data backtesting, essentially training its models to think the way expert forecasters do, then checking their work against what actually happened.

The team behind the predictions

Mantic was founded in 2024 by Toby Shevlane, a former Google DeepMind scientist, and Ben Day, who holds a machine learning PhD from Cambridge. The team also includes alumni from DeepMind and major financial institutions like Citadel.

Before this seed round, Mantic had raised approximately $4M in pre-seed funding in 2025. The company had already demonstrated state-of-the-art accuracy on Metaculus benchmarks before its cup victory.

Why medium-term forecasting matters

There’s a reason Mantic specifically targets medium-term predictions. Short-term forecasting is already well-served by traditional statistical models. Long-term forecasting is so uncertain that even the best models produce outputs too vague to act on. The medium term is the sweet spot where decisions actually get made and where traditional methods tend to struggle the most.

The company says it’s already seeing interest from both enterprise clients and government agencies globally. The $25M will go toward scaling these capabilities, improving model accuracy, and expanding the team.