Skip to content

Manifold

📋Page Status
Page Type:ContentStyle Guide →Standard knowledge base article
Quality:43 (Adequate)⚠️
Importance:18.5 (Peripheral)
Last edited:2026-01-29 (3 days ago)
Words:4.1k
Structure:
📊 24📈 2🔗 0📚 6817%Score: 13/15
LLM Summary:Manifold is a play-money prediction market with millions of predictions and ~2,000 peak daily users, showing AGI by 2030 at ~60% vs Metaculus ~45%. Platform scored Brier 0.0342 on 2024 election (vs Polymarket's 0.0296), demonstrating play-money markets can approach real-money accuracy but with systematic gaps due to lower liquidity and financial incentives.
Issues (2):
  • QualityRated 43 but structure suggests 87 (underrated by 44 points)
  • Links3 links could use <R> components
DimensionAssessmentEvidence
ScaleLargeMillions of predictions, thousands of active users
AI CoverageExtensiveDedicated AI categories, timeline markets
AccuracyGoodCompetitive with Polymarket on most questions
AccessibilityHighFree to use, user-created markets
CommunityActiveManifest conference, EA/rationalist engagement
Funding$1.84M+FTX Future Fund, SFF, ACX Grants
Business ModelEvolvingPlay money core; sunset real-money feature 2025
AttributeDetails
Full NameManifold (formerly Manifold Markets)
FoundedDecember 2021
FoundersAustin Chen, James Grugett, Stephen Grugett (CEO)
LocationSan Francisco Bay Area / Austin, Texas
StatusFor-profit startup
Websitemanifold.markets
CurrencyMana (play money)
Key EventManifest Conference (annual)
GitHubmanifoldmarkets/manifold
InvestorsLeonis Capital, Soma Capital

Manifold is an online prediction market platform where users engage in competitive forecasting using play money called “Mana.” Unlike traditional forecasting platforms that require question approval, Manifold allows anyone to create markets on any topic, enabling rapid coverage of emerging events and niche questions. This permissionless approach has made Manifold particularly valuable for AI forecasting, where new developments constantly create new prediction opportunities.

Founded in December 2021 by Austin Chen and brothers James and Stephen Grugett, Manifold emerged from the effective altruism and rationalist communities with seed funding from the Astral Codex Ten grant program. The platform quickly grew to become one of the most active prediction markets, with particular strength in technology, AI, and intellectual community questions. By March 2024, Manifold had reached over 2,000 daily active users and up to 200,000 unique visitors per month, though user activity declined in 2025 following the sunset of real-money features.

Manifold’s design emphasizes social features and accessibility over maximum accuracy. Users can follow forecasters, comment on markets, and build reputations through the platform’s leagues system. While this approach sacrifices some accuracy compared to real-money markets like Polymarket, it dramatically lowers barriers to participation and enables forecasting on questions too niche or legally risky for real-money platforms. The platform has become an unofficial “training ground” for professional traders, with many top-ranked individuals on Kalshi having started by accumulating Mana on Manifold.

Austin Chen co-founded Manifold in December 2021 alongside the Grugett brothers. Chen studied at UC Berkeley and built a career in technology before founding Manifold, having worked as an engineer at Google and as a Tech Lead at Streamlit. He also previously founded One Word. Chen brought experience in both large tech companies and startups to Manifold’s development.

In April 2024, Chen departed from Manifold to focus full-time on Manifund, the related charitable giving platform. His farewell post cited four reasons for leaving: Manifold had become stable with less room for iteration, he wasn’t excited about the planned pivot to real-money sweepstakes, prediction markets seemed “insufficiently powerful” for his interests, and short AI timelines were reshaping his priorities. Chen’s wife, Rachel Weinberg, is a software engineer who joined Manifund as co-founder and engineer.

Stephen Grugett serves as CEO of Manifold. He grew up in Atlanta with his brother James and developed early economic intuitions playing games like RuneScape and Neopets, where trading was central. Grugett studied computer science and philosophy at Yale, then worked as a programmer at SIG (Susquehanna International Group), an options trading firm, for a year.

After leaving SIG, Grugett moved to New York to work at a friend’s robo-advisor startup for a year, then became a digital nomad living primarily in Southeast Asia. During this period, he and James started their previous venture—a subscription group chat app for online creators—before co-founding Manifold.

James Grugett is a co-founder of Manifold and attended Carnegie Mellon University. He worked closely with his brother Stephen on both their earlier subscription group chat startup and Manifold. James has been an active speaker at Manifest conferences, delivering talks including “Prediction Markets Will Save the World” at Manifest 2023.

In late 2024, James participated in Y Combinator’s Fall 2024 batch with a new project called Manicode (now Codebuff), a CLI tool that uses AI to write code with your codebase as context. While James has shifted some focus to this new venture, he remains involved with Manifold.

Loading diagram...
PeriodKey Events
Dec 2021Founded by Austin Chen, James Grugett, and Stephen Grugett with ACX Grants seed funding
2022Received 1.5M USD from FTX Future Fund (before collapse); 340K USD+ from SFF
Sept 2023First Manifest conference with 250 attendees including Nate Silver, Robin Hanson
Apr 2024Austin Chen departs to focus on Manifund
June 2024Manifest 2024 grows to 600 attendees
Nov 2024Major coverage of 2024 US election
Mar 2025Sweepcash real-money feature sunset to refocus on core platform
2025User activity declines; record low of 886 daily active traders in March
Loading diagram...
TypeDescriptionUse Case
BinaryYes/No questions”Will GPT-5 be released by 2025?”
Multiple ChoiceSeveral options”Which lab will release AGI first?”
NumericNumber predictions”What will GPT-5’s MMLU score be?”
DateWhen something happens”When will AI beat humans at X?”
Free ResponseOpen-ended optionsExploratory questions
CurrencyTypeStatus
ManaPlay moneyPrimary currency, free to earn
SweepcashReal moneySunset March 28, 2025
Prize PointsRedeemableFor charity donations

Manifold distinguishes itself from other prediction markets through its emphasis on social features. The platform functions as a “social prediction game” designed for community engagement rather than pure financial speculation.

Leagues System: Users compete in seasonal leagues that rank forecasters by performance. “Mana Whales”—users who have accumulated large balances through accurate forecasting—hold significant status within the community and often use their wealth to “boost” niche markets on topics like science and AI safety.

Reputation Capital: Users develop track records measured by Brier Score, which they can use to build credibility. Market creators who demonstrate reliability earn followers, positive reviews, and increased activity on their questions. This reputation system has made Manifold an effective training ground for professional forecasters.

Incentive Structure:

  • Daily streak bonuses provide escalating rewards (M5 on day one, increasing by M5 daily to a M25 cap)
  • Quests and bounties offer targeted Mana payouts for tasks like market creation
  • In EA communities, Mana earnings can be redeemed for donations to approved charities

Topics and Groups: Users can filter markets by topic, create custom groups, and invite communities from other platforms. Popular topics include AI, politics, sports, and effective altruism.

Manifold is open source and provides a comprehensive API for developers. The platform’s data architecture includes:

ComponentTechnologyPurpose
Public APIVercel-hostedMarket listing, search, basic operations
DatabaseSupabase (SQL)Primary data storage (migrated from Firebase)
Internal APIGoogle Cloud DockerComplex operations (e.g., share trading)
Python LibrarymanifoldpyData analysis, accuracy metrics, API bindings

The API supports market creation, trading operations, liquidity management, and information retrieval. Developers use it for probability modeling, sentiment analysis, and research projects. Type definitions are available in the repository’s common/src/api/schema.ts.

In 2024, Manifold launched “Sweepcash,” a real-money betting feature where users could redeem winnings for cash (with a 5% fee) or donate to charity (no fee). The experiment aimed to increase user engagement and provide stronger forecasting incentives.

However, in March 2025, Manifold announced the sunset of Sweepcash, effective March 28, 2025:

ReasonExplanation
Usage goals not metReal-money trading didn’t attract expected user growth
Platform focus dilutedSweepstakes development drew resources from core features
Regulatory complexityOperating real-money markets required ongoing compliance work
Strategic realignmentDecision to “double-down on user-created markets that make Manifold unique”

After the sunset, non-cashed-out Sweepcash was converted to Mana at a rate of 100:1. Despite losing real-money features, Manifold has emphasized that users continue trading for “reputation capital” and forecasting skill development rather than financial returns.

Manifold has become a significant venue for AI forecasting, with an active AI market dashboard and dedicated AGI timelines page. The platform’s permissionless market creation enables rapid coverage of AI developments that regulated markets cannot address.

Question TypeExamplesActivity Level
AGI Timelines”AGI by 2028?”, “AGI by 2030?”Very High
Model Releases”GPT-5 release date”, “Claude 4 capabilities”High
Benchmarks”ARC-AGI grand prize”, “Turing Test by 2029”High
Lab Events”Will Anthropic raise X?”, “OpenAI leadership”Moderate
Safety Events”AI safety incident by 2030?”Moderate
Superintelligence”Superintelligence by 2030?”, “by 2040?”Moderate
QuestionManifoldMetaculusPolymarket/Kalshi
AGI by 2028≈47%≈30%
AGI by 2030 (OpenAI)High volume YES40% (Kalshi)
AGI by 2030≈60%≈45%9% by 2027 (Polymarket)
Weak AGI by end 20253%
Superintelligence by 203026%
Superintelligence by 204069%
ARC-AGI grand prize78%
Turing Test (Long Bets) by 202950-53%

Manifold’s AGI estimates tend to be higher than Metaculus, partly due to different user bases, less stringent AGI definitions, and the platform’s rationalist community overlap. A combined AGI Timelines Dashboard aggregating multiple sources estimates AGI will arrive in 2031 (with 80% confidence range of 2027-2045) as of January 2026.

The 2024 US presidential election provided a significant test of prediction market accuracy. Analysis of platform performance revealed clear differences between play-money and real-money markets.

Brier Score Comparison (2024 Presidential Election)

Section titled “Brier Score Comparison (2024 Presidential Election)”
PlatformMean Brier ScoreCurrency TypeNotes
Polymarket0.0296Real (crypto)Best performer
Futuur0.0315Real
Kalshi0.0319Real (USD)
Manifold0.0342Play moneyStatistical difference from Polymarket
Nate Silver0.0396N/ALeast accurate

All prediction markets scored below 0.035, reflecting high accuracy overall. However, a two-tailed paired t-test found Polymarket beat Manifold with statistical significance (p < 0.05).

MetricManifoldPolymarketWinner
Overall AccuracyGoodBetterPolymarket
High-Volume RacesCompetitiveSuperiorPolymarket (liquidity advantage)
Market StabilityMore stable throughoutMore volatileManifold
Trump ConfidenceLowerHigherPolymarket (more accurate)
Democrat Wins (Gallego, Slotkin)Less accurateMore accuratePolymarket
  1. Trading Volume: Polymarket had billions in volume vs. much lower activity on Manifold
  2. Real vs. Play Money: Financial incentives on Polymarket drove more rigorous forecasting
  3. Arbitrage: Real-money markets can be arbitraged against each other; play-money cannot
  4. Marginal Corrections: Lower liquidity on Manifold made it slower to adjust prices by a few percentage points

Despite the gap, Manifold was “pretty close” to Polymarket, and the difference was primarily at the margins. Both platforms outperformed traditional polls, consistent with broader findings that prediction markets beat polls in forecasting elections.

Platform Calibration and Accuracy Research

Section titled “Platform Calibration and Accuracy Research”

Manifold publishes real-time calibration data that assesses whether events occurred as often as predicted. The methodology samples 2% of all past trades on resolved binary questions with 15 or more traders every hour.

MetricValueNotes
Sample Size≈89,000 tradesAs of late 2024
Sampling Rate2% of all tradesHourly updates
Minimum Traders15 per marketFor inclusion in calibration
Optimal Traders10-20+Research shows calibration stops improving beyond this
FindingSourceImplication
2022 Midterms: Manifold outperformed real-money markets, nearly matched FiveThirtyEightManifold internal analysisPlay-money can compete with real-money in some contexts
16+ traders sufficient for quality predictionsAcademic researchSmaller markets may exhibit favourite-longshot bias
Play-money vs. real-money accuracy: No systematic difference across 208 games”Prediction Markets: Does Money Matter?” studyTheoretical justification for play-money approach
Favourite-longshot bias exists; predictions improve closer to eventPage & Clemen researchLong-term forecasts biased toward 50%
Metaculus vs. Manifold Brier scores: 0.111 vs. 0.168Comparative analysisCurated platforms may outperform permissionless ones

Calibration City, a cross-platform accuracy analysis project funded through Manifund, integrates data from Kalshi, Manifold, Metaculus, and Polymarket with over 130,000 total markets.

SourceAmountYearNotes
Astral Codex Ten GrantsSeed funding2021Initial support
FTX Future Fund1,500,000 USD2022Before FTX collapse
Survival and Flourishing Fund340,000+ USD2022+Ongoing support
Leonis CapitalUndisclosed2022+Institutional investor
Soma CapitalUndisclosed2022+Institutional investor
Total Known1,840,000+ USDGrants and investments

Manifold hosts “Manifest,” an annual festival for forecasts, markets, and ideas. It represents the largest in-person gathering of the prediction market and forecasting community, attracting participants from EA, rationalist, economics, journalism, and tech communities.

YearDatesAttendanceKey Features
Manifest 2023Sept 22-24250First in-person forecasting festival; held at Lighthaven
Manifest 2024June 7-9600Expanded program; part of 10-day extravaganza with LessOnline
Manifest 2025June 6-8TBDAnnounced; registration open

The inaugural Manifest was held September 22-24, 2023 at Lighthaven in Berkeley. Sponsors included Kalshi, Polymarket, Sovereign, and Metaculus.

Sample Schedule:

  • 10-11 AM: Opening session
  • 11-12 PM: Fireside chat with Robin Hanson
  • 1-2 PM: Estimathon (fermi estimation with prizes)
  • 2-3 PM: Speed friending
  • 4-5 PM: Panel: Forecasting Founders (Manifold, Metaculus, Polymarket, Kalshi, FRI)
  • 7-8 PM: Workshop: How to Write Good Forecasting Questions
  • 8 PM-12 AM: “Murder She Bet” (murder mystery + prediction market game)

Manifest 2024 was held June 7-9, 2024, part of a 10-day event series with Lightcone Infrastructure’s LessOnline conference the preceding weekend.

Programming Innovations:

  • Interest-specific meetups for politics, journalism, AI, mechanism design
  • Serious talks combined with attendee-run workshops
  • Fun side events including prediction-market-based games
  • Overnight accommodations at Lighthaven
SpeakerBackgroundYear(s)
Nate SilverFiveThirtyEight founder, The Signal and the Noise author2023, 2024
Scott AlexanderAstral Codex Ten author2023, 2024
Robin HansonEconomist, prediction market pioneer2023, 2024
Dwarkesh PatelPodcaster, interviewer2024
Cate HallPoker player, forecaster2024
Eliezer YudkowskyAI safety researcher, MIRI founder2023
Dylan MatthewsVox journalist2023
Zvi MowshowitzForecaster, blogger2023
Katja GraceAI Impacts founder2023
AellaResearcher, data analyst2023
DestinyPolitical streamer2023
Robert MilesAI safety communicator2023
Richard HananiaPolitical scientist2023
Emmett ShearFormer Twitch CEO, brief OpenAI CEO2023
James GrugettManifold co-founder2023
CategoryExample Sessions
Prediction Markets & JournalismScott Alexander (ACX) + Dylan Matthews (Vox)
Markets vs. Financial MarketsByrne Hobart (The Diff)
Genetic EnhancementJonathan Anomaly
Platform ComparisonsFounders from Manifold, Kalshi, Polymarket, Metaculus
Quantitative ToolsSquiggle and other forecasting tools

Manifund is a charitable giving platform co-founded by Austin Chen and closely related to Manifold (hence the similar names). The platform experiments with innovative funding mechanisms to support effective altruism and AI safety projects.

AttributeDetails
Founded2023
Co-foundersAustin Chen, Rachel Weinberg
BoardAustin Chen, Barak Gila, Vishal Maini
StaffRachel Weinberg (full-time), Austin Chen (~half-time as of 2023)
2023 Funding Distributed2,000,000+ USD
2025 Regrant Budget2,250,000 USD
ProgramDescription2023-2024 Scale
RegrantingDonors delegate budgets to expert “regrantors” who make fast, small grants1.6M USD to 6 AI safety regrantors in 2024
Impact CertificatesRetroactive funding based on demonstrated impact≈50K USD invested in 12 projects via ACX 2024
Open ApplicationsDirect applications for fundingPart of 2M USD total 2023
Prize ChallengesCompetition-based fundingExperimental in 2024

Manifund’s signature innovation is its regranting model. Rather than centralized grantmaking, donors give to individual “regrantors”—experts in specific fields—who then distribute funds according to their judgment.

Key Advantages:

  • Speed: Manifund can move dollars to grantees within days
  • Expertise: Regrantors have domain knowledge donors may lack
  • Autonomy: Regrantors can fund time-sensitive opportunities (compute, travel, talent)

2024-2025 AI Safety Regrantors (100K USD+ budgets each):

  • Adam Gleave (Redwood Research)
  • Dan Hendrycks (CAIS)
  • Evan Hubinger (Anthropic researcher)
  • Leopold Aschenbrenner (former OpenAI, “Situational Awareness” author)
  • Ryan Kidd
  • Neel Nanda (mechanistic interpretability researcher)

For 2025, Manifund raised 2.25M USD and announced their first 10 regrantors.

Manifund has experimented with impact certificates—a mechanism for retroactive funding where investors purchase shares in projects and later sell them to donors who want to fund demonstrated impact.

2024 ACX Grants Partnership:

  • ACX directly funded 33 projects for 1.35M USD
  • Impact market funded 12 additional projects for ~50K USD at ~200K USD combined valuation
  • Projects that opted-in but didn’t receive direct funding could seek impact market investment

Lessons Learned:

  • Difficult to attract sufficient investor interest
  • Haven’t found use cases where certificates led to clearly better decisions than traditional grants
  • Show promise but haven’t achieved product-market fit
  • Need more participants and bigger use cases to validate
PlatformCurrencyMarket CreationStrengthsWeaknesses
ManifoldPlay moneyAnyoneAccessibility, breadthLower accuracy
PolymarketCrypto (real)ApprovedAccuracy, liquidityLegal restrictions
KalshiUSD (real)ApprovedRegulated, legitimateLimited topics
MetaculusReputationApprovedAI focus, rigorNot a market
PredictItUSD (real)ApprovedPolitical focusCapped positions
MetricValuePeriodNotes
Total Markets Created100,000+CumulativeUser-created, permissionless
Peak Daily Active Users≈2,000March 2024Before Sweepcash sunset
Monthly Unique VisitorsUp to 200,000March 2024Peak engagement
2025 Daily Active Traders≈886-1,000March-July 2025Record low in March 2025
User Decline≈50%Jan 2024 - Mar 2025Daily active traders roughly halved
AI/Tech Markets5,000+CumulativeHigh activity category
Total PredictionsMillionsCumulativeAcross all market types

The platform experienced significant growth through 2023-2024, peaking around the 2024 US election. However, user activity has declined substantially in 2025, with the number of daily active traders roughly halving since records began in early 2024. The March 2025 low of 886 daily active traders represents a new record low.

Contributing factors to the decline include:

  • Sunset of Sweepcash real-money feature
  • Reduced novelty after initial growth phase
  • Competition from expanding real-money platforms like Kalshi
  • Questions about long-term business model sustainability

The prediction market industry has grown significantly:

MetricLate 2025
Total Industry Volume≈$13 billion/month
Key MilestoneKalshi won CFTC lawsuit (Oct 2024)
Regulatory TrendIncreasing legitimacy

Manifold occupies the “accessible, social” niche in this growing ecosystem.

StrengthEvidence
AccessibilityFree to use, no financial risk, anyone can create markets on any topic
SpeedNew markets can respond to events within minutes of news breaking
BreadthCovers niche topics other platforms cannot (legal, regulatory, or liquidity constraints)
CommunityActive discussion, social features, strong EA/rationalist engagement
InnovationExperiments with new market types (date markets, numeric), leagues, embedding
Training GroundMany professional forecasters developed skills on Manifold before moving to real-money platforms
Open SourceTransparent codebase; developers can audit and contribute
LimitationEvidence
Lower AccuracyPlay money reduces financial incentives; Brier score gap vs. Polymarket (0.0342 vs. 0.0296)
LiquidityLower trading volume makes markets slower to adjust to new information
Manipulation RiskEasier to move markets without financial cost
Business Model UncertaintyRevenue model unclear after Sweepcash sunset; depends on grants
User Decline50% drop in daily active traders from 2024 peak to 2025
Regulatory RiskPrediction market legal status continues evolving

Manifold serves several unique functions in AI forecasting that complement other platforms:

  1. Rapid Coverage: When new AI developments occur, Manifold markets often appear within hours
  2. Niche Questions: Topics too specific or speculative for curated platforms
  3. Community Thermometer: Reflects EA/rationalist community beliefs, which influence AI policy discourse
  4. Training Data: Large corpus of resolved forecasts useful for calibration research

However, for high-stakes decisions, real-money markets (Polymarket, Kalshi) or curated platforms (Metaculus) may provide more reliable signals.