Mantic raises $25m to sell its tournament-tested forecasts to traders

On 18 September, Mantic, a London-based company that builds AI systems to forecast world events, announced a $25m (£18.7m) funding round led by Radical Ventures. Balderton, Thinking Machines, DRW, FT Ventures, M12 (Microsoft’s venture capital fund), Episode 1, Charlie Songhurst and Thomas Wolf also took part, according to the company’s announcement. Reuters, which interviewed the company’s chief executive, reported the round as seed funding at an undisclosed valuation.

The round follows a contest result: in the summer 2026 Metaculus Cup, an online tournament in which entrants put probabilities on around 60 world events, Mantic said its system beat all 676 human participants, professional forecasters included. Only one entrant finished ahead of it, another bot called laertes, according to the results as reported by Reuters. In the fall 2025 cup the same company ranked fourth of 539 human forecasters, which its website then described as the best AI result to date.

Mantic Technologies was incorporated on 11 July 2024, Companies House records show. Its co-founders are Dr Toby Shevlane, the chief executive, and Dr Ben Day, the chief technology officer, and each holds more than 25% but not more than 50% of the shares. Shevlane was previously a research scientist at Google DeepMind. He told Reuters the idea came from that job, where he needed help predicting global events relevant to AI. The company came out of stealth in 2025 with a $4m (£3m) pre-seed round led by Episode 1 with backing from DRW, according to its launch post.

The contest result leads Mantic’s announcement, but the buyers it names first are not tournament organizers. Its first market is firms that trade on events, where a forecast is judged against a price within days. The corporate planning market it also claims is slower to judge and, so far, unnamed. The $25m is a bet that a points lead over human forecasters turns into decisions people will pay for.

How the forecasts are built

Mantic builds on frontier models from other labs. Shevlane told Reuters that the company specializes those models to be better at forecasting, testing its system on historical data, grading the results and then improving it. The announcement said models such as GPT-6 and Kimi K3 “provide a strong foundation, but need the right scaffolding and fine-tuning” to forecast well. In the same cup, Mantic said, it beat the other AI startups and every frontier model entered by the organizers on a standard harness.

The product built on that system lets users create and track their own forecasts on the issues that affect their work, with access through an API and the Model Context Protocol (MCP). The company said it beat the crowd on questions ranging from geopolitics and technology to Spider-Man box office revenues.

“We’re now upgrading the level at which humans can understand the future,” Shevlane told Reuters.

Traders first, then the boardroom

Mantic said that some of the world’s leading hedge funds already use it to predict market-moving events in geopolitics, macroeconomics and business, and that it has done especially well at predicting when events will happen. It also said Fortune 500 customers are adopting it to inform decisions on M&A and new product releases. It named none of them, and it declined to name customers when Reuters asked.

Aaron Rosenberg, the Radical Ventures partner who has joined Mantic’s board, said hedge funds and trading firms were particularly eager for its forecasts, Reuters reported. Companies House lists Rosenberg as a director since 22 June 2026. DRW, which describes itself as “a diversified trading firm”, backed both the pre-seed and the new round.

The announcement sets out where the money goes: a significantly larger team and more compute and data behind the forecasts. Mantic said it is hiring across AI research, engineering, product development, sales and operations, and that it is onboarding new customers, “especially traders in financial markets”.

Trading firms also judge a supplier faster than corporate buyers do. A trading desk can test a probability against a market move and drop the supplier within a quarter if the numbers don’t hold. A forecast behind an acquisition or a product launch may take a year or more to prove right or wrong, and the buyer rarely publishes the result.

What the round has to prove

The corporate case so far rests on unnamed customers and inbound interest. Mantic said it receives inbound interest across applications “from supply chain management and demand forecasting to executive-level planning”. Those are the decisions a chief executive or finance director at a mid-market firm would recognize, and none of them has yet been attached to a customer Mantic is willing to name.

The contest record is public and checkable, which is unusual for an AI company at seed stage. Metaculus scores every entrant on the same questions, and Mantic’s ranking has improved from fourth among humans in fall 2025 to ahead of all of them in summer 2026. The company said in its announcement: “Only last year, experts believed this achievement was still a decade away.”

A tournament, though, is scored on questions the organizers pick. A hedge fund scores Mantic on the questions it trades, and a corporate buyer on the decisions it makes. The $25m pays for the team and compute to be tested on those terms, and until a customer goes on the record, the only published scorecard Mantic has is the one Metaculus keeps.