The Pulse
Mantic Raises $25 Million After AI Beats Human Forecasters
London AI startup Mantic raises a $25 million seed round led by Radical Ventures after finishing ahead of every human participant in the Summer 2026 Metaculus Cup. The company specializes frontier AI models for probabilistic forecasts about
AI.info Team ·
Mantic finishes ahead of every human participant in the Summer 2026 Metaculus Cup, then raises $25 million to turn that result into a business. The London startup says its system assigned probabilities to political, economic and cultural events more accurately than the human competitors in the online tournament, which concluded in September.
Radical Ventures leads the seed round at an undisclosed valuation. Microsoft’s venture fund M12, Thinking Machines Lab, Balderton Capital and other investors also participate, according to Reuters.
A tournament result puts Mantic ahead of humans
The Metaculus Cup tests forecasters on questions that resolve against real-world outcomes. Mantic beats every human contestant and all but one bot, named laertes, according to the tournament results cited by Reuters. The performance marks the first time technology dominates the competition, Reuters reports.
Mantic’s result does not establish that its system predicts every event better than people. It shows that the company performed best among the human and AI participants in that particular competition, using its scoring rules and question set.
The company’s public materials describe forecasts covering one week to one year across geopolitics, business, policy, technology and culture. Mantic presents its system as a way to generate probability estimates, supporting reasoning and references rather than a single categorical answer.
Radical Ventures leads the $25 million round
Radical Ventures partner Aaron Rosenberg joins Mantic’s board as part of the investment. He tells Reuters that companies and government agencies around the world have shown interest, with some already integrating Mantic’s AI. Mantic does not identify those customers.
Hedge funds and trading firms are among the strongest early sources of demand, Rosenberg says. “If Mantic is superhuman as it is, they can make money off of that immediately,” he tells Reuters.
The round gives Mantic a larger pool of capital after its earlier financing. The company’s website says it previously raised $4 million in pre-seed funding led by Episode 1, with participation from the trading firm DRW and angel investors connected to Google DeepMind and Anthropic.
Specialized models, historical tests and probability scores
Mantic does not train a foundation model from scratch for the forecasting task. CEO Toby Shevlane tells Reuters that the company specializes frontier models developed by other laboratories, tests the resulting system against historical data, grades its performance and then improves it.
Shevlane co-founded Mantic with Ben Day in 2024. Mantic’s website describes Shevlane as a former senior research scientist at Google DeepMind and Day as its chief technology officer and co-founder.
Shevlane says his experience at Google DeepMind helped prompt the company after he needed better tools for predicting global events relevant to AI. The company’s approach is aimed at questions where data alone is insufficient and researchers must gather context, weigh competing signals and update probabilities as new information arrives.
Examples from music charts to Colombia’s election
Shevlane tells Reuters that Mantic avoided a consensus error on a Billboard Hot 100 question involving Shakira’s “Dai Dai” and “Waka Waka.” Human forecasters largely expected “Dai Dai” not to overtake “Waka Waka,” but that prediction proved wrong. Mantic did not place as much weight on the prevailing view, Shevlane says.
In another example, Mantic gave Colombian presidential candidate Abelardo De La Espriella about a 40% chance of winning, compared with roughly 30% for the consensus forecast. De La Espriella ultimately won, according to Reuters.
Those examples illustrate the company’s pitch: avoid simply copying a crowd’s estimate, search broadly for relevant evidence and preserve enough uncertainty to account for less popular outcomes. The tournament result gives that pitch a measurable public benchmark, while the undisclosed customer list leaves commercial performance harder to assess.
Funding turns a public benchmark into a commercial test
Mantic now has to show that its tournament performance carries into decisions made outside a competition. Investors are backing a system designed for business, finance and government users, but the company has not disclosed customer names, contract sizes or financial results.
The immediate evidence remains the Summer 2026 Metaculus Cup: Mantic ranks above every human forecaster and below one competing bot. Its new $25 million round, led by Radical Ventures, puts money behind the next test—whether probabilistic forecasts can improve decisions when the questions are proprietary, the information is incomplete and the outcome carries a financial or policy cost.