Key points

  • Mantic raised $25 million in a seed round led by Radical Ventures at an undisclosed valuation.
  • Microsoft's M12, Thinking Machines Lab and Balderton Capital were among the participating investors.
  • The company's system beat every human entrant and finished behind one other bot in the summer 2026 Metaculus Cup.

London-based artificial-intelligence startup Mantic has raised $25 million in seed funding after its forecasting system outperformed every human entrant in the summer 2026 Metaculus Cup. Founder and chief executive Toby Shevlane announced the financing on September 18, saying the company was created to improve predictions of world events. Reuters independently reported that Radical Ventures led the round at an undisclosed valuation, with participation from Microsoft's M12 venture fund, Thinking Machines Lab, Balderton Capital and other investors.

Funding follows a measurable test

The financing arrives with a benchmark that is unusually concrete for an early-stage AI company. Reuters said Mantic assigned probabilities to political, economic and cultural outcomes more accurately than all human competitors in the latest Metaculus tournament, finishing behind only another automated entrant called laertes. Mantic's website says the system has improved across four consecutive tournaments, moving from fourth place among 539 human forecasters in summer 2025 to first among humans in summer 2026. Competition performance does not prove accuracy in every commercial setting, but it gives investors an external record to evaluate.

Related reporting: Nscale files for NYSE IPO after $140.6M first-half revenue

How the forecasting system works

Mantic does not describe itself as a builder of a new frontier foundation model. Instead, it adapts models developed by other laboratories for forecasting tasks, tests their predictions against historical events and uses the resulting scores to improve the system. Shevlane told Reuters that the approach grew from his earlier work at Google DeepMind, where he needed better forecasts of global developments relevant to AI. The company says its goal is to provide probability estimates that businesses and governments can use when decisions depend on uncertain future conditions.

Investors target decision-making markets

Radical Ventures partner Aaron Rosenberg has joined Mantic's board, according to Reuters. He said companies and government agencies have shown interest in the technology, while hedge funds and trading firms represent an obvious potential market because improved forecasts can have direct financial value. Mantic has not disclosed customer names, revenue or the valuation attached to the round. Those omissions limit an outside assessment of current adoption, even as the investor roster signals institutional interest in forecasting as a distinct AI category.

A test of AI beyond text generation

Forecasting systems must combine current information, historical comparisons and calibrated uncertainty rather than produce a single confident answer. That makes tournament scoring important: probabilities can be evaluated after events resolve, allowing systems to be compared over time. Mantic says its technology is designed to reduce dependence on group consensus, while Reuters reported examples in which its probabilities differed materially from human forecasts. The larger challenge will be translating tournament performance into reliable decisions when data are incomplete, incentives differ and rare events carry high costs.

What the new capital changes

The $25 million round gives Mantic resources to develop its forecasting platform and pursue commercial deployments without the company having to claim that prediction is solved. The disclosed facts are narrower: a fresh seed financing, a strong external competition result and an investor group spanning specialist AI capital and major technology companies. Future evidence will come from repeated benchmark performance, named deployments and measurable outcomes for customers. For now, the deal marks a sizable bet that probabilistic forecasting can become a practical enterprise product rather than remain a research exercise.

Sources

AI-generated editorial image; not a photograph of the reported event. Prepared with AI assistance and source verification.