---
title: "Mantic Raises $25M Seed After Beating Humans at Metaculus Cup"
description: "Radical led Mantic’s $25M seed with Balderton, M12, and Thinking Machines Lab after the London forecasting startup beat every human in the Summer 2026 Metaculus Cup — who buys calibrated AI forecasts and why."
date: 2026-09-18T00:00:00.000Z
source: https://venturecapitaltracker.com/2026-mantic-25m-seed-radical-balderton-forecasting
---

# Mantic Raises $25M Seed After Beating Humans at Metaculus Cup

> Radical led Mantic’s $25M seed with Balderton, M12, and Thinking Machines Lab after the London forecasting startup beat every human in the Summer 2026 Metaculus Cup — who buys calibrated AI forecasts and why.

**[Mantic](/startup/mantic)** raised **$25 million** in seed funding announced **September 18, 2026**, led by **Radical Ventures**. **Microsoft M12**, **Thinking Machines Lab**, and **[Balderton Capital](/fund/balderton-capital)** participated. Valuation **undisclosed**. Radical’s Aaron Rosenberg joined the board.

AI that is merely fluent is cheap; AI that is **correct about what happens next** is scarce — and Mantic just posted a public tournament score that beat every human.

## Key facts

| Field | Detail |
| --- | --- |
| Company | [Mantic](/startup/mantic) — London; founded 2024 |
| Founders | Toby Shevlane (ex-DeepMind), Ben Day |
| Round | **$25M** seed |
| Lead | Radical Ventures |
| Participants | M12, Thinking Machines Lab, Balderton, others |
| Valuation | **Not disclosed** |
| Proof point | Summer 2026 Metaculus Cup — ahead of all humans; one AI (laertes) finished higher |
| Customers | Interest from companies, governments, hedge funds (Rosenberg) — **names not disclosed** |
| Model strategy | Adapts third-party frontier models; does not train its own foundation model from scratch |

## Who uses the product — and for what job

**Users (claimed interest):** hedge funds and trading desks, corporate strategy teams, and government agencies that already buy human forecasting or research.

**Job:** assign **calibrated probabilities** to messy political, economic, business, and cultural questions weeks or months out — where crowd consensus fails and chatbots hallucinate with confidence. Tech Startups cites examples where Mantic broke from crowd error (Shakira chart race; Colombia election probabilities).

Rosenberg’s line is blunt: if the system is superhuman, buyers “can make money off of that immediately.” That is a **decision-support** sale, not a seat license for writing email.

## Why now

- Metaculus Cup 2026 marked the first time **AI systems dominated** the field (Reuters) — a category-shift signal, not a marketing benchmark buried in a paper.
- Labs are shipping general agents; forecasting is a narrower, **scoreable** job where wrong answers have P&L consequences.
- Cap table mixes a Toronto AI specialist (Radical), a European multi-stage firm ([Balderton](/fund/balderton-capital)), Microsoft’s M12, and Thinking Machines Lab — research credibility plus commercial distribution hypotheses.

## Why Radical / Balderton — portfolio fit

**Radical Ventures** leads AI-native companies and put a partner on the board — classic seed leadership when the product is research-heavy. **[Balderton Capital](/fund/balderton-capital)** is London’s European multi-stage franchise; a UK forecasting lab fits its “European founders, global ambition” pattern even at seed ticket size. M12 adds distribution into enterprise Microsoft ecosystems. Thinking Machines Lab adds frontier-lab adjacency without being the lead check.

**

| Investor | Fit |
| --- | --- |
| Radical | AI thesis lead; board seat |
| Balderton | European founder coverage; multi-stage follow-on path |
| M12 | Enterprise / Microsoft channel optionality |
| Thinking Machines Lab | Research signal |

**Strategic rationale:** take an AI-specialist lead that will help productize research, keep Balderton for European growth later, and use lab-adjacent angels for credibility with the same buyers who read Metaculus leaderboards.

## Competitive map

- **Human forecasting platforms** (Metaculus, Good Judgment) — tournaments and crowds; Mantic competes as AI entrant and commercial product.
- **Generic LLMs** — text fluency ≠ calibrated probabilities scored against outcomes.
- **Quant / alternative-data shops** — structured markets; Mantic targets judgmental, event-driven questions.

## What remains undisclosed

- Valuation and ownership.
- Paying customer names, ARR, and contract sizes.
- How performance transfers outside Metaculus question types and horizons.
- Dependency risk on third-party frontier model APIs.

## Implication

If tournament edge becomes commercial edge, Mantic is a bet that **the next AI budget line is being right**, not being chatty. One Cup does not prove durable alpha — the open question is whether hedge-fund and policy buyers pay for continuously scored forecasts once the leaderboard news cycle fades.

### Sources

1. Reuters (Sep 18, 2026): https://www.reuters.com/technology/ai-startup-mantic-raises-25-million-superhuman-forecasting-2026-09-18/
2. Tech Startups: https://techstartups.com/2026/09/18/british-ai-startup-mantic-raises-25m-to-build-superhuman-ai-forecasting-after-metaculus-win/

**By:** [Venture Capital Tracker](https://venturecapitaltracker.com/editorial-policy)
**Last updated:** September 19, 2026

**Editorial note:** AI tools assisted with research, structure, or drafting. Venture Capital Tracker retains human editorial responsibility for factual accuracy, relevance, and source quality before publication.