# Manifold

Anyone can open a question, anyone can take a side, and the currency buys nothing.

*https://predictionmarkets.tools/tools/manifold · Forecasting Platforms*

## Facts

### At a glance

| Field | Value |
| --- | --- |
| Vendor | Manifold |
| Category | Forecasting Platforms |
| Job | forecasting |
| Website | https://manifold.markets |
| Pricing model | one-time |
| Free tier | true |
| Open source | true |
| Licence | MIT |
| Self-hosted | false |
| Tested hands-on | false |
| Last updated | 2026-09-21 |

### Pricing

| Tier | USD | Period |
| --- | --- | --- |
| Mana pack | 5 USD | — |
| Membership | on request | — |

### Availability

| Field | Value |
| --- | --- |
| Jurisdictions | global |
| Open to US persons | true |
| KYC required | false |

### Markets

| Field | Value |
| --- | --- |
| Subjects | politics, sports, macro, crypto, science, culture, business |
| Settlement | play_money |
| Resolved by | Whoever created the question resolves it; moderators overturn or re-resolve only in exceptional circumstances, and only moderators and staff can annul a question to N/A. |

### Economics

| Field | Value |
| --- | --- |
| Taker fee | none |
| Maker fee | none |
| Liquidity model | amm |
| Platforms | web, ios, android |
| AI features | none |

### Interfaces

| Field | Value |
| --- | --- |
| API | true |
| WebSocket | true |
| Scripting | none |
| Python | false |
| MCP server | false |
| Export | json, api |

### Capabilities

Yes: charting, order_book, automation, live_trading, portfolio_tracking, calibration_scoring, alerts

No: screening, backtesting, paper_trading, cross_venue, news, tax_reporting

*Verified: pricing 2026-09-19; availability 2026-09-19; markets 2026-09-19; economics 2026-09-19.*

## What it is

A prediction market where anyone can open a question in about thirty seconds, anyone can take a
side of it, and the person who opened it decides how it resolves. Everything is denominated in
mana, written M, which Manifold's own documentation describes in four words as play-money currency
that cannot be converted to cash.

That is the whole product and also the whole argument about it. Removing money removes the
gatekeeping: there is no deposit, no licence, no market-listing committee, and consequently there
are tens of thousands of questions on subjects no exchange would ever list. It also removes the
mechanism everyone cites as the reason a market price means anything. Manifold's answer is to
publish its own calibration curve — on 19 September 2026 the platform-wide figure on its
calibration page was a Brier score of 0.17375 — and to argue in public that the number holds up.

**Read the denominator, because the page publishes one and it is not "every market".** The method
described beside the figure samples 2% of all past trades every hour, on resolved binary questions
with fifteen or more traders, and the stated sample was 98,000 trades. It is trade-weighted rather
than time-weighted, which Manifold notes itself, adding that accuracy may be better than the chart
shows because large miscalibrated trades are usually corrected immediately. The page does not say
whether the Brier figure covers that same sample or something wider. Read on 21 September 2026.

**That figure has no unit printed on it, and Brier has two conventions in circulation.** One sums
the squared error over both outcomes of a binary question and is bounded at 2; the other scores a
single outcome and is bounded at 1. They differ by a factor of two on every number, so 0.17375
means two quite different things depending on which Manifold used, and its calibration page does
not say. Read against the definition this catalogue uses on the
[forecasting platforms page](https://predictionmarkets.tools/categories/forecasting-platforms), which is the two-outcome one, it
would be a strong number; read the other way it is an ordinary one. Do not rank it against a score
from another platform without settling that first.

Pricing is an automated market maker: you trade against a curve rather than against a queue of
other people's orders, with limit orders resting on top of it that other traders fill when they
push the probability through your level. Creating a question subsidises its own liquidity pool out
of the creation cost.

### What mana actually is

Worth being precise, because this is the field readers get wrong. Mana is bought from Manifold at
a flat 100 mana per US dollar — the smallest credit-card pack is $5 for M500 — and it flows the
other way not at all. The terms say it has no cash or redeemable value and cannot be cashed out,
sold, traded, transferred, gifted or exchanged, on the site or elsewhere. There is no longer a
charity-donation route out either.

So the honest description is a consumable, not a currency. You can spend mana on positions, on
creating questions, on subsidising someone else's liquidity, and on Manifold's own digital goods.
You cannot get a dollar back out of it, and a leaderboard denominated in it is a leaderboard of
skill at Manifold, which overlaps with but is not the same as being right.

Loans complicate this slightly. A free daily loan against your open positions is interest-free and
available to everyone; a margin loan, which requires a paid membership, accrues 0.03% interest per
day, is deducted from your proceeds when a position closes whether or not your balance covers it,
and can take your balance negative. In mana, but negative.

### The sweepcash episode

Between September 2024 and March 2025 Manifold ran a parallel sweepstakes currency, sweepcash,
which eligible users could redeem for cash. It is gone. Manifold announced the wind-down on 13
February 2025, halted new sweepcash issuance from purchases, signups and streaks immediately,
resolved sweepstakes markets closing after 3 March to their current probability, gave users until
28 March to cash out or donate, and converted everything left at 100 mana to 1 sweepcash. The
stated reason was that sweepstakes had not met usage goals and were pulling attention away from
the core platform. Anything you read describing Manifold as a real-money venue is describing that
five-month window.

## Availability

Open worldwide, free to join, and no identity check to use it. The terms are the usual
California-company shape — the site is not tailored to any jurisdiction, and anyone accessing it
from elsewhere does so on their own initiative. Minimum age is 13, with parental permission
required for minors.

Two places where identity verification does appear, and neither is a condition of forecasting.
Verifying identity unlocks a larger starting mana balance and a referral bonus. And Manifold runs
prize drawings with real prizes, which carry their own geographic eligibility rules, no-purchase
terms, and an identity check through a third-party provider before a prize is paid. Read the
drawing's own rules rather than assuming the platform's openness carries over to them.

## Pricing

Free, with two things for sale, neither required.

Mana packs are the one with a published number, and it is published in Manifold's own repository:
a flat 100 mana per US dollar on credit-card purchases with no purchase bonus, in packs of $5,
$25, $100 and $1,000. Since mana cannot come back out, this is a purchase and not a deposit.

Memberships — Plus, Pro and Premium — are the second, and they are what gates margin loans and
higher leverage. They are not bought with dollars: the shop configuration in Manifold's repository
prices them at M500, M2,500 and M10,000 for thirty days, alongside quest-reward multipliers and a
shop discount. Mana is the only thing here you can spend money on.

Trading the question markets is free. The taker fee constant and the flat trade fee in the
open-source code are both zero, which is the answer the play-money design forces anyway. The
perpetuals are the exception and the one place a fee exists: opening a position costs 10 basis
points of notional on all 27 live perp markets on 19 September 2026, closing costs nothing, and
the fee goes into that market's backing pool rather than to Manifold. Trades authenticated with an
API key are charged separately, 10 to 30 basis points across those markets — a separate rate the
code introduces after two episodes in August 2026 in which bots trading a stale oracle mark drained
the Bitcoin perp pools.

## Markets & resolution

Every subject, because anybody can list anything: elections, sports, macro and equity prices, AI
capability and safety, awards and reality television, personal accountability questions, and a
large tail of questions about Manifold itself. Market types include binary, multiple choice with
independent or summing answers, numeric, polls, bounty questions, and perpetuals — leveraged long
or short positions on a live number with no expiry, priced off an oracle feed rather than off a
crowd, which is a different product wearing the same currency.

Resolution is the part to understand before trusting a price here. **The creator resolves their
own question.** Manifold says so plainly and treats it as a design choice, encouraging clear
criteria written in advance and stepping in only in exceptional circumstances. Moderators can
overturn a resolution, and re-resolve for abuse, misresolution or technical failure, after a user
report. Resolving to N/A cancels a question and is restricted to moderators and staff. There is
also a PARTIAL resolution, where the creator picks a percentage rather than an outcome.

Compare that with a venue, where resolution is the operator's obligation against a named source
and money is sitting on the answer, and with [Metaculus](https://predictionmarkets.tools/tools/metaculus), where staff resolve
against published criteria and annul what cannot be evaluated. On Manifold the resolver is a
stranger with no stake in being consistent, which is the standing risk on any Manifold price and
the reason the creator's track record is worth a look before taking a side.

## Integrations

The API is genuinely public in the way Metaculus's no longer is: read endpoints need no
authentication, writes need a key, and the documented rate limit is 500 requests per minute per
IP. There is a real-time websocket server for market, trade and other event updates, which needs a
ping every 30 to 60 seconds or the connection is dropped. Bots and automated trading are
explicitly permitted; scraping outside the API is not.

Free bulk dumps of bets, markets and comments are offered as zipped JSON for personal and
non-commercial use — and they are stale. The data page states the dumps were last updated on 6
July 2024. Anything more recent, or any commercial or model-training use, goes through a data
licence.

The platform is MIT-licensed and the repository was last pushed to on 18 September 2026. Manifold
does not offer it as a self-hosted product, and nothing in the docs describes running your own
instance.

## Limitations

Creator resolution is the one to weigh. A market is only as good as the person who wrote it, and
the recourse when they get it wrong is a report to a moderator rather than a rulebook.

The public data dumps have not moved since July 2024, so a research project that needs history
either lives on the API and its rate limit or pays for a licence.

Mana profit is the headline number on a profile, and profit is not accuracy. Someone who trades
many small mispricings in thin markets and someone who is well calibrated on hard questions end up
in very different places on a mana leaderboard, and the leaderboard cannot tell you which you are
looking at. The platform-wide Brier score is a claim about Manifold as an aggregate, not about any
individual on it.

The perpetuals are a second product on the same balance and they do not behave like the rest of
the site. Positions carry leverage — capped at 3x on eighteen of the 27 live markets, 10x on six
and 100x on three — they can be liquidated, they price off an oracle feed rather than off the
crowd, and they are the only thing here that charges a fee. A calibration argument about Manifold's
question markets says nothing about them.

And the currency does not leave. Buying mana is buying a consumable; the terms are unusually blunt
about that, and it is the right way round to read them.

## Alternatives

[Metaculus](https://predictionmarkets.tools/tools/metaculus) if you want a proper scoring rule and a record that means something
outside the site — fewer questions, slower, scored rather than traded.
[Good Judgment Open](https://predictionmarkets.tools/tools/good-judgment-open) if you want to be ranked against a crowd on
geopolitical questions by Brier score, with a route to professional work at the end of it.

## FAQ

### Can mana be cashed out?

No. Manifold's terms state that mana has no cash or redeemable value and cannot be cashed out, sold, traded, transferred, gifted or exchanged for money, property, products or services, on the site or anywhere else. It can be bought, and it can be spent on the platform.

### Does Manifold still have real-money markets?

No. The sweepcash layer that ran from September 2024 was closed in 2025 — Manifold announced the wind-down on 13 February 2025, stopped issuing sweepcash immediately, gave users until 28 March to redeem, and converted the remainder to mana at 100 mana per sweepcash.

### What does it cost to trade on Manifold?

Nothing on the question markets — the taker fee constant and the flat trade fee in Manifold's own open-source code are both zero. Opening a perpetual position costs 10 basis points of notional, paid into that market's backing pool rather than to Manifold; closing is free. There is no money in the system to take a cut of either way, and creating a question costs mana, which goes into that question's liquidity.

### Why is Manifold in two categories here?

Because it is genuinely hired for two jobs. People come to Manifold to be right about something and watch a track record, and they also come to take a side against somebody who disagrees. Nothing else in this catalogue is in both halves of that sentence.

## Also from Manifold

- [Manifold API](https://predictionmarkets.tools/tools/manifold-api.md)

## Background

- [How accurate a prediction market's price actually is](https://predictionmarkets.tools/guides/how-accurate-the-price-is.md) — What research establishes about reading 73 cents as a 73% chance — where market prices are calibrated, the deviations that recur, and what the horizon does.
- [What a forecasting platform's score actually measures](https://predictionmarkets.tools/guides/what-a-forecasting-score-measures.md) — Proper rules, the two Brier conventions that differ by a factor of two, crowd-relative scores that do not travel between sites, and why profit is not accuracy.
- [What a prediction market is, and how one works](https://predictionmarkets.tools/guides/what-a-prediction-market-is.md) — A contract that pays one dollar if a stated event happens. What you buy, who takes the other side, who decides, and when the cents stop meaning a probability.

## Also worth comparing

- [Metaculus](https://predictionmarkets.tools/tools/metaculus.md) — Proper scoring and public track records on questions nobody can take a position in.
- [Prediki](https://predictionmarkets.tools/tools/prediki.md) — A wiki with a market attached — anyone writes the question, edits it, votes the result.
- [Social Science Prediction Platform](https://predictionmarkets.tools/tools/social-science-prediction-platform.md) — Forecast what a study will find, before it finds it, and be scored against the estimate.
- [Confido](https://predictionmarkets.tools/tools/confido.md) — Open-source forecasting workspace you host yourself, with a Brier score pointing upwards.
- [Fatebook](https://predictionmarkets.tools/tools/fatebook.md) — Write down what you think will happen, in Slack or a browser, and get scored on it.
- [Good Judgment Open](https://predictionmarkets.tools/tools/good-judgment-open.md) — Brier scores against the crowd, and the recruiting ground for Superforecasters.

## Named as a replacement for

- [Augur](https://predictionmarkets.tools/alternatives/augur.md)
- [INFER](https://predictionmarkets.tools/alternatives/infer.md)
- [Kalshi](https://predictionmarkets.tools/alternatives/kalshi-alternatives.md)
- [PredictionBook](https://predictionmarkets.tools/alternatives/predictionbook.md)
- [Trepa](https://predictionmarkets.tools/alternatives/trepa.md)
- [Zeitgeist](https://predictionmarkets.tools/alternatives/zeitgeist.md)