Mantic Raises $25 Million After Its AI Beats Human Forecasters in a Tournament

Mantic Raises $25 Million After Its AI Beats Human Forecasters in a Tournament

September 22, 2026

London-based startup Mantic has raised $25 million after its system outperformed human participants in the 2026 Metaculus Cup, Reuters reports. The competition included 676 human forecasters and questions about political, economic, and cultural developments. The result attracted interest from investors and organizations that want better ways to reason about uncertain events.

Mantic says it specializes existing AI models with additional data and forecasting methods. That matters because a general language model’s plausible explanation is not the same as a calibrated probability. A useful forecaster must express uncertainty consistently and improve when evidence changes.

What a tournament result shows

Forecasting tournaments provide structured questions with outcomes that can later be checked. A scoring rule rewards probabilities that are closer to reality, making it possible to compare performance over many questions. But a single competition cannot establish that a system will forecast every domain well.

Results depend on which questions are included, how long predictions remain open, and whether the evaluation is representative of a real user’s decisions. Organizations should examine calibration, performance by topic, update behavior, and failure cases—not only the overall rank.

Mantic’s own announcement describes the model’s performance and commercial interest. Reuters independently reported the funding and competition result, while noting the company’s claims about customer use. Those claims should be distinguished from independently audited outcomes.

The value and risk of AI forecasts

Forecasting tools can help businesses plan supply chains, assess product launches, or monitor geopolitical risks. A system can rapidly synthesize public information and generate a probability estimate. It can also amplify stale data, miss context, or produce false confidence if users mistake a number for certainty.

Decision-makers should treat forecasts as inputs rather than instructions. Good practice includes recording the question, probability, timestamp, evidence, and the decision informed by the forecast. Comparing predictions with outcomes helps reveal whether the tool is actually useful.

What to watch

The company’s next challenge is proving performance outside a tournament. Watch for long-term calibration records, transparent scoring, coverage across domains, and documentation of how the system updates. If customers make high-stakes decisions from forecasts, they should require human review and a clear account of uncertainty.

Mantic’s result is an interesting signal in a growing field. AI may help people make better predictions, but the standard is not whether a forecast sounds persuasive. It is whether probabilities are consistently accurate and improve decisions over time.

Sources

• Reuters: Mantic raises $25 million for AI forecasting
• Mantic: Funding and Metaculus Cup results
• Metaculus