Hypermind

What it is

A free prediction market that has been running since May 2014, organised as a series of parallel contests with real prize money and no deposit. You register, you are handed 200,000 ℍ of virtual currency once and for all — there are no top-ups — and you trade shares of outcomes against other forecasters. A correct outcome's shares expire at 100 ℍ and everything else at 0 ℍ, so a price of 80 ℍ reads as an 80% probability.

Two companies stand behind the name, and the card names both because the terms do. The corporate site is published by Hypermind SAS of Neuilly-sur-Seine, France, with Emile Servan-Schreiber as publication director; the game itself is organised by Hypermind llc, a New York entity, and the rules are governed by Connecticut law. The platform is the descendant of NewsFutures, which is where the academic literature on this crowd starts.

What ranks you here is trading profit, not a scoring rule. Your performance in a market contest is the sum of the square roots of your profits on each question minus the sum of the square roots of your losses — the square root is there so that the same total profit scores better when extracted from many questions than from one. There is no Brier score and no log score on the leaderboard. That is a real difference from the rest of this category, and it means a Hypermind rank measures skill at this market rather than calibration in the abstract.

A second mode, Prescience, drops the order book: you assign probabilities directly, every update is recorded daily, and each is scored on proximity to the answer and against what most other participants said. Prescience points accumulate within one contest and are not transferable between contests — there is no cumulative career score across the two modes.

On 19 September 2026 the contests page showed 24,573 € at stake in total, of which 4,000 € sat on the general Prediction Markets contest, alongside a contest called Glimt described as a Swedish government project developed with Ukraine and run on Hypermind. Open markets that day included control of the US Senate after the November 2026 elections and the succession to Macron.

Availability

Open worldwide in principle and free to enter, with three conditions worth reading before you spend a season on it. You must be at least 21 at registration. The rules state the game is void where prohibited or restricted by law, without naming a country, so the enforcement is a term rather than a geoblock. And the registration form asks for profile information that Hypermind reserves the right to refuse to activate if it doubts the information is truthful or complete.

There is no identity check to sign up and nothing to deposit. Identity comes up once, at the other end: the rules say a winner may be required to furnish proof of identity, address and birth date in order to receive a prize, and that there can be only one winner per family and postal address.

The rule most likely to bite a casual reader is inactivity. A forecaster who has not bought or sold a share or entered a Prescience forecast for four consecutive weeks is deactivated by notice; a balance of earnings can still be cashed for one month after that, and is voided afterwards. Separately, a global performance score — the sum of your last five closed contests — is what Hypermind uses to identify under-performing forecasters who may be removed from the panel. This is a panel with a cull, not an open forum.

Pricing

Free to enter, and there is no transaction fee of any kind: when a buy order matches a sell order, the buyer pays the price and the seller receives it, with nothing taken out. Prizes run the other way. Each contest carries a fixed reward in euro or dollars set by whoever sponsored it — public or private organisations commissioning forecasts on their own uncertainties — and the pot is divided pro rata of performance among the forecasters who ended with a positive performance. Twice the performance is twice the share; a negative performance pays nothing.

Earnings accumulate across contests and are paid out as Amazon gift certificates, not as a bank transfer. The two pages that state the payout minimum disagree with each other: article 8 of the rules sets it at 25 €, while the site's How It Works page says the minimum is 20 USD or 15 euro. Read the rules if the difference matters to you. Hypermind's own July 2026 analysis says 72,000 € has been distributed across the markets it covers, which is twelve years of prize money on 1,141 resolved markets.

Taxes on a prize are the winner's problem, and the rules say so.

Markets & resolution

Subject matter is elections and geopolitics first, then economics and business, with climate and technology questions alongside. Sport is present but thin, and where it appears it is usually a multinomial tournament-winner question rather than a single match. That is a deliberate difference from a cash venue, where the sports questions are the volume.

Resolution is Hypermind's, and the rules are unusually explicit that it is discretionary. Hypermind may halt a question — cancelling resting orders while keeping the position open — close it when the correct outcome is observed, or void it outright. A voided question cancels every trade in it along with the profits and losses; a conditional question whose condition never happens is voided the same way. Where the correct outcome cannot be established, Hypermind may instead pay the outcomes at intermediate prices, for example 50/50 on a yes/no question. The rules close the section by saying the organiser is the sole judge of the moment to halt, close or void, and that the payout decision is final.

Multi-outcome questions allow pre-selling: Hypermind lends you shares against a deposit of 100 ℍ each, you sell them, and you buy them back later — or have them bought back for you at expiry.

Integrations

There is no API, no websocket feed and no SDK. Nothing here is built to be automated against, and the panel rules point the other way: the questions, their outcomes, their prices and the forum are described as strictly confidential intellectual property that a participant may not share outside the panel.

Against that sits the one genuinely useful machine-readable artefact in this category. Alongside its accuracy report of 14 July 2026, Hypermind published its full trading history as two files: a 44 MB CSV of every trade — timestamp, question id, outcome, price, quantity — beginning on 16 May 2014, and a plain-text file of winning outcomes per question. That is 981,180 trades across 1,141 resolved markets, and it is the raw material for checking the platform's accuracy claims yourself rather than taking them. It is a dump and not a feed: the date is baked into the filename, and nothing refreshes it.

Limitations

The leaderboard is a profit leaderboard. If what you want is a calibration record you can quote in a research context, a rank here does not give you one, and there is no per-forecaster Brier or log score to lift.

The accuracy figures are Hypermind's own analysis of Hypermind's own data. They are checkable, because the data is published — the report puts the binary skill score at 50.2%, the multinomial at 38.4% and overall at 45.6%, and claims parity on calibration with two named cash venues — but the analysis, the matching methodology and the choice of comparison set are all the vendor's.

The confidentiality clause is the awkward one. It forbids sharing question text, listed outcomes and prices with anyone outside the panel, which rules out quoting a live Hypermind probability in public and sits oddly beside the historical dump the same company publishes.

Activity is mandatory. Four weeks without a trade deactivates the account, and repeated poor global performance is grounds for removal from the panel. Payment is in gift certificates only. And the platform is small by the standards of the cash venues in this catalogue — twelve years of trading comes to under a million trades, which is a quiet week elsewhere.

Alternatives

Metaculus is the same audience scored properly: a log score with two published rescalings, a public record, and nothing to trade. Good Judgment Open is free, Brier-scored against the crowd median, and its prize is a route to paid work rather than money. Manifold keeps the order book and the play currency but lets anyone open and resolve a question, with no prize pool and no panel to be removed from.