LONDON — An artificial intelligence startup that recently outperformed human forecasters in predicting real-world events has secured $25 million in seed funding, putting a major bet behind a technology designed to turn AI into a more accurate engine for anticipating what comes next.
London-based Mantic announced the funding on Sept. 18, with Radical Ventures leading the round. Investors also include Microsoft’s venture fund M12, Thinking Machines Lab and Balderton Capital, according to Reuters.
The investment follows Mantic’s performance in the Summer 2026 Metaculus Cup, an online forecasting tournament that tested predictions across political, economic and cultural events. Metaculus says the summer tournament closed in early September, with winners announced Sept. 5.
From AI experiment to forecasting business
Mantic was co-founded in 2024 by Toby Shevlane and Ben Day. Shevlane previously worked as a research scientist at Google DeepMind, where the challenge of predicting developments relevant to artificial intelligence helped inspire the company.
Rather than building an entirely new frontier model, Mantic specializes existing AI models and designs a forecasting system around them.
The startup’s approach involves testing predictions against historical outcomes, measuring performance and repeatedly refining the system. The goal is not simply to generate an answer, but to produce probabilities that can be compared with what actually happens.
That distinction is important in forecasting: an AI system can be evaluated after the event, giving developers a measurable feedback loop for improving its predictions.
The result that caught investors’ attention
Reuters reported that Mantic’s system assigned probabilities to political, economic and cultural developments more accurately than the human competitors in the Summer 2026 Metaculus Cup.
The result represents a significant change from Mantic’s earlier tournament performance.
In the 2025 Metaculus Cup, Mantic finished eighth, a result that was already notable because it placed an AI system among the competition’s leading forecasters. Contemporary reporting at the time also emphasized that the strongest human forecasters continued to outperform AI in some settings.
Metaculus’ own platform now lists its Summer 2026 Cup as closed, with 58 questions and winners announced in September.
The broader forecasting landscape is still developing, however. Metaculus research published earlier in 2026 said professional human forecasters had continued to outperform individual bots in several controlled comparisons, underscoring why a single tournament result should not be interpreted as proof that AI has universally surpassed humans at forecasting.
Why AI forecasting is attracting attention
Mantic’s proposition goes beyond predicting sporting results or trivia.
The company is targeting questions involving geopolitics, business, technology, policy and culture—areas where organizations routinely make decisions despite uncertainty.
One example cited by Reuters involved Colombia’s presidential election. Mantic reportedly assigned candidate Abelardo De La Espriella roughly a 40% probability of winning early in the tournament, compared with approximately 30% for the human consensus at the time. De La Espriella ultimately won.
The system also reportedly diverged from human consensus over the performance of Shakira’s song “Dai Dai” relative to “Waka Waka” on the Billboard Hot 100. The human consensus expected “Dai Dai” not to overtake the older hit, while Mantic did not make the same heavily consensus-based assumption.
Mantic’s chief executive has argued that avoiding the tendency of humans to cluster around a prevailing consensus can be an advantage when forecasting uncertain events.
Microsoft and major investors are taking notice
The new funding round gives Mantic backing from several prominent technology and venture investors.
M12, Microsoft’s venture investment arm, participated in the financing, while Thinking Machines Lab and Balderton Capital also invested, Reuters reported. Radical Ventures led the round.
Sky News had reported a day earlier that Mantic was raising more than $20 million, with the round jointly led by Balderton and Radical Ventures and M12 participating. Sky News also reported that sources estimated the company’s valuation at close to $90 million, although that valuation was not part of the company’s publicly announced funding figure and should therefore be treated as reported rather than confirmed.
The discrepancy in timing is significant: the Sky News report described the fundraising while it was still being assembled, while the subsequent Reuters report reflects Mantic’s announced $25 million financing.
Hedge funds and governments could become major customers
The commercial opportunity may be even more significant than the tournament result.
Aaron Rosenberg, a Radical Ventures partner who joined Mantic’s board, told Reuters that companies and government agencies around the world have expressed interest in the technology, with some already integrating Mantic’s AI.
The company has not disclosed the identities of those customers.
Rosenberg said hedge funds and trading firms were particularly interested because better probability estimates could potentially be incorporated into financial decision-making.
That does not mean Mantic can reliably predict markets or geopolitical events in every situation. Rather, the company’s pitch is that consistently better-calibrated probabilities could provide organizations with another source of information when making decisions under uncertainty.
The bigger question: Can AI actually forecast the future better?
Mantic’s progress arrives as researchers and technology companies increasingly treat forecasting as a measurable AI capability.
The appeal is straightforward: unlike many AI benchmarks, forecasts eventually have real-world outcomes. A prediction can be recorded, scored and compared with competing forecasts.
But forecasting remains difficult.
Human experts bring domain knowledge, intuition and the ability to recognize unusual circumstances. AI systems, meanwhile, can process large amounts of information, run repeatedly without fatigue and update predictions as new information arrives.
The evidence so far suggests the answer is not as simple as “AI has replaced humans.”
The 2025 Metaculus competition, for example, showed Mantic breaking into the top ranks while human experts still retained advantages in some forecasting situations. Earlier 2026 research from Metaculus likewise found professional forecasters ahead of individual bots in controlled comparisons.
What has changed is the speed at which AI forecasting systems are improving.
Mantic’s rise—from eighth place in the 2025 Metaculus Cup to a much stronger showing in the 2026 competition—has helped turn forecasting from an experimental AI application into a potentially valuable commercial market.
And with $25 million in new funding, Mantic now has considerably more resources to find out whether a machine can do something humans have struggled with for centuries:
not just understand what is happening now—but put a number on what happens next.
