Mantic Raises $25M After Beating Human Forecasters

Forecasting earns its keep when someone must act before certainty arrives. A probability has to influence a trade, product launch, acquisition plan, supply decision, or policy choice while the outcome is still capable of making everyone involved look foolish. Mantic has raised $25M to build for that uncomfortable interval. The London AI forecasting company announced the financing on September 18, 2026, after its system finished ahead of all 676 human entrants in the Summer 2026 Metaculus Cup. Radical Ventures led the round, with Balderton Capital, Thinking Machines Lab, DRW, FT Ventures, M12, Episode 1, Charlie Songhurst, and Thomas Wolf participating.

Reuters described the financing as a seed round and reported that the valuation was not disclosed. Mantic previously raised a $4M pre-seed, bringing its disclosed funding across the 2 rounds to $29M.

What Mantic Is Building

Mantic specializes in what it calls judgmental forecasting: predicting events that require current research, contextual reasoning, and calibrated probabilities rather than a clean historical dataset with an obvious pattern. Its questions can span geopolitics, macroeconomics, business, technology, and culture.

The company's product allows customers to create and track personalized forecasts, with MCP and API access for incorporating those probabilities into other workflows. CEO Toby Shevlane told Reuters, in reporting syndicated by CNA, that Mantic specializes frontier models from other AI labs for forecasting, tests the system on historical questions with known outcomes, scores the results, and improves from that feedback.

That approach treats forecasting as a discipline rather than a confident paragraph about the future. A useful system has to express uncertainty, update when evidence changes, and receive a score that punishes misplaced confidence. The product claim is therefore measurable in a way many AI demonstrations are not.

The Metaculus Result and Its Limits

The Summer 2026 Metaculus Cup gave Mantic its sharpest public credential. The system finished ahead of all 676 human participants, including professional forecasters, and behind one bot called laertes. Mantic also says it beat every other AI startup and the frontier models entered by organizers through a standardized harness.

The distinction matters. Mantic did not prove that software now predicts every market, election, conflict, or business outcome better than every person. It produced the strongest human-comparison result inside a defined tournament, across that event's questions, scoring rules, timing, entrants, and resolution process.

That is still meaningful. Forecasting tournaments make systems commit probabilities before answers are known. They expose calibration, reward useful disagreement, and create a record that can be scored after reality arrives. Mantic says it separated from the human crowd on questions including Colombia's presidential election, where it assigned greater probability to the eventual winner earlier than the community did.

Why Investors Put $25M Behind the Result

The commercial opportunity sits between a forecast and the decision it changes. Mantic says hedge funds already use its system to assess market-moving events across geopolitics, macroeconomics, and business. It also says Fortune 500 customers use forecasts to inform M&A and product-launch planning.

Reuters reported that companies and government agencies around the world have integrated Mantic, citing Radical Ventures partner and new Mantic board member Aaron Rosenberg. The customers remain unnamed, and the public record does not disclose contract values, revenue, retention, or measured decision outcomes. Those gaps keep the commercial claims in their proper category: early adoption evidence, not an audited business-performance record.

Radical Ventures is funding a company whose product can be evaluated before the market is fully formed. The tournament supplies a visible accuracy signal. The participation of M12, Thinking Machines Lab, DRW, FT Ventures, Balderton Capital, and returning investor Episode 1 adds institutions with direct exposure to AI research, enterprise software, trading, media, and early-stage company building.

The Founders' Route Into Forecasting

Mantic was founded in London in 2024 by Toby Shevlane and Ben Day. Shevlane, the company's CEO, previously spent 2.5 years as a Senior Research Scientist at Google DeepMind and worked on experiments testing Gemini's dual-use capabilities. His Oxford doctoral work examined governance decisions around releasing large language models and managing misuse risk.

Day, Mantic's CTO, previously led research at Foresight Data Machines, where he developed AI for steel-production optimization. He holds a Cambridge PhD in machine learning focused on meta-learning and graph neural networks and also worked on AI for drug development at Relation Therapeutics.

Their backgrounds point toward the product's central problem. Decision-makers rarely need another system that sounds certain. They need a system that can search across incomplete evidence, expose its reasoning, assign a probability, and improve when the world proves it wrong.

What the New Capital Changes

In its funding announcement, Mantic said the $25M will expand the team and increase the compute and data powering its forecasts. The company is hiring across AI research, engineering, product development, sales, and operations while onboarding more customers, especially traders in financial markets.

Scaling the research system is only part of the work. Mantic must preserve calibration as models, data sources, question types, and customer contexts change. It also has to make forecasts legible enough that a portfolio manager, corporate strategist, or public-sector leader can understand why the probability moved and decide whether the change deserves action.

The Summer Cup gave Mantic a result that can survive more scrutiny than a polished demo. The next layer of evidence will come from repeated tournaments, transparent error analysis, independent evaluation, and customer decisions that can be traced from forecast to outcome. The $25M gives Mantic more capacity to build that record while its probabilities are moving closer to the rooms where capital, policy, and operating plans get committed.