Mantic Raises $25 Million After Beating Human Forecasters

Mantic’s AI system beat every human competitor in a recent forecasting tournament, yet the startup has not named the organizations using it. The London company has now raised $25 million in seed funding after finishing ahead of all but one bot in the 2026 Metaculus Cup.

Radical Ventures led the round at an undisclosed valuation. Microsoft’s venture fund M12, Thinking Machines Lab and Balderton Capital also participated, and Radical partner Aaron Rosenberg joined Mantic’s board.

The contest result behind the pitch

The summer Metaculus Cup asked participants to assign probabilities to political, economic and cultural developments. Mantic placed behind only a bot called laertes; Reuters described the event as the first competition in which AI entrants dominated the field.

Toby Shevlane, Mantic’s chief executive, co-founded the company with Ben Day in 2024. Shevlane previously worked as a research scientist at Google DeepMind, where he said a need to forecast AI-relevant global events helped inspire the company.

A system built on other labs’ models

Mantic does not present itself as the developer of a new frontier model. Instead, Shevlane said, it specializes frontier models from other labs for forecasting, evaluates them against historical data, grades their performance and iteratively improves the system.

Mantic’s stated improvement cycle

• Adapt frontier models from other labs for forecasting.
• Test forecasts against historical data and grade the results.
• Use those evaluations to improve the system over time.

A claimed edge against consensus

Shevlane said part of Mantic’s advantage was avoiding herd mentality. He cited a question on whether Shakira’s “Dai Dai” would overtake “Waka Waka” on the Billboard Hot 100: human forecasters heavily backed the wrong outcome, while Mantic did not place a heavy bet on that consensus.

He also said Mantic gave Abelardo De La Espriella about a 40% chance of winning Colombia’s presidential election early in the tournament, against a roughly 30% consensus forecast. De La Espriella ultimately won. Those examples are Shevlane’s account of the system’s performance.

Interest is clearer than the customer list

Rosenberg said companies and government agencies around the world had integrated Mantic’s AI, though Mantic declined to identify customers. He said hedge funds and trading firms were particularly interested in its forecasts, making the company’s contest showing a useful signal—but not a public record of commercial results.