Skip to content

Numerai

Free tournament

AI hedge fund that crowdsources ML models from data scientists worldwide through encrypted tournaments

What is Numerai?

Numerai is an AI hedge fund founded by Richard Craib in 2015 with a radical thesis: instead of hiring in-house quants, pay thousands of anonymous data scientists to train machine learning models on encrypted stock market data, then ensemble the best predictions into a meta-model that trades real capital. Participants download a regularly-updated dataset of obfuscated features, train any ML model they like, and submit predictions each week. Models that perform well on live markets earn NMR (Numerai's Ethereum token), while models that perform poorly lose staked NMR. Numerai's mechanism design gives data scientists skin in the game and gives Numerai a diverse set of uncorrelated models. In addition to the main Numerai Tournament, the company runs Numerai Signals (where participants submit predictions on raw, non-encrypted tickers) and the Numerai Meta Model (the live trading ensemble). Numerai is free to join and free to submit — users only need NMR to stake. With over 6,000 active data scientists and $100M+ in staked NMR at various points, Numerai is the largest AI-driven crowdsourced quant fund in the world. The platform has become a popular proving ground for ML engineers who want to test models on real financial data without running their own fund.

⚡ Quick Verdict

Best for

Data scientists and ML engineers who want to apply their skills to live markets

Not ideal for

Passive investors looking to allocate capital to a managed fund

Starting price

Free to participate · NMR staking optional

Free plan

Yes — full tournament access is free, staking is optional

Key strength

Novel mechanism design that rewards ML model performance with cryptocurrency stake

Limitation

Encrypted features mean you can't do traditional financial analysis — pure ML only

Bottom line: Scores 4.3/5 — A legitimate place for ML practitioners to test models on real markets. Most stakers lose money, so treat it as skill-building, not income.

Pricing

Tournament participation — Free: Download the weekly dataset, train any model, and submit predictions. No cost, no subscription.

Staking — NMR required: To earn (or lose) on your model's live performance, stake NMR tokens. Minimum stake depends on the tournament. NMR is an ERC-20 token traded on Coinbase, Binance, and other exchanges. Rewards and burns happen weekly based on correlation and meta-model contribution.

Numerai Fund (accredited investors only): A separate hedge fund vehicle — not available through the public site.

Key Features

  • Weekly encrypted dataset covering thousands of global equities
  • Unlimited free model submissions with no hardware minimum
  • NMR staking — earn or burn based on live market performance
  • Numerai Signals for submitting predictions on raw tickers
  • Open leaderboards showing correlation, TC, and rewards
  • Public forums, Discord, and Kaggle-style community
  • Python API (numerapi) for automated submission pipelines
  • Numerai Meta Model — the live ensemble trading real capital

Pros & Cons

Pros

  • Completely free to start — no fees, no AUM minimums
  • Real feedback on ML models from live market performance
  • Vibrant community with public notebooks and tutorials
  • Novel, transparent mechanism design with on-chain rewards

Cons

  • Most stakers lose money — the tournament is highly competitive
  • Encrypted features remove domain intuition; pure ML only
  • NMR token volatility adds non-model risk to payouts
✅ Pricing verified April 2026 · ✅ Independently reviewed · ✅ Scoring methodology

FAQ

Is Numerai a real hedge fund or just a game?

Both. The Numerai Tournament is a public ML competition where data scientists submit predictions and stake NMR on their models. Behind the scenes, Numerai Global Equity runs a real hedge fund that combines the best submitted models into a meta-model and trades equities at institutional scale. The fund has been live since 2016. Retail participants don't invest in the fund directly — they compete in the tournament for NMR rewards.

How much can you earn on Numerai?

Realistically, most data scientists lose money staking NMR, especially when starting out. Top performers can earn double-digit percentage yields on their stake during good periods, but burns are equally common in bad periods. Treat Numerai as a skill-building and experimentation platform, not an income source. The payout structure is designed to reward long-term model quality.

What kind of ML models work best on Numerai?

Gradient boosted trees (XGBoost, LightGBM, CatBoost) have historically dominated, with neural networks and ensembles gaining ground. The community publishes public notebooks and baselines that routinely place in the top half of the leaderboard. The real differentiator is feature engineering, era-aware training, and managing the correlation-to-meta-model trade-off. Creativity and robustness beat raw compute.

Is NMR (Numeraire) a good investment?

NMR is a utility token required for staking — not a conventional investment. Its price is driven by tournament demand, token burns, speculation, and broader crypto cycles. Numerai periodically burns NMR from poorly performing stakes, which creates deflationary pressure, but prices remain volatile. Treat NMR exposure separately from your view of Numerai as a company.

How is Numerai's encrypted data different from normal market data?

Instead of raw fundamentals or prices, Numerai provides hundreds of numerical features with meaningless names (feature_1, feature_2, ...) and a target that's some transformation of future returns. The idea is to remove survivorship bias and prevent participants from leaking outside data. You can't apply traditional finance intuition — you're forced to do pure statistical learning.

Is Numerai legal in the US?

Yes. The tournament is free to enter and NMR staking is framed as performance-based rewards, not an investment security. Numerai is based in San Francisco. That said, NMR is an ERC-20 token and US users should check their state tax rules — staking rewards and burns may be taxable events. Consult a tax advisor for specifics.

What's the difference between the Tournament and Signals?

The main Tournament uses Numerai's encrypted, era-based dataset — you train on obfuscated features. Numerai Signals lets you submit predictions on real, raw tickers using any data you bring yourself (fundamentals, alt data, price history). Signals is more flexible but also more competitive since participants can use whatever edge they have. Both contribute to the live meta-model.

📋 Good to know

Setup

Create a free account at numer.ai, download the dataset, and submit your first predictions with numerapi. No credit card needed.

Privacy

Numerai is a public tournament. Model submissions are private but leaderboard performance is visible under your username.

When to stake

Only stake NMR after your model has at least 20 weeks of live, out-of-sample performance. Never stake more than you can afford to burn.

Learning curve

Medium to high — requires Python, ML basics, and comfort with time-series cross-validation.

Explore more

Compare Numerai with alternatives

Numerai vs KenshoFull comparison →Numerai vs AlphaSenseFull comparison →Numerai vs Stock UnlockFull comparison →Numerai vs Simply Wall StFull comparison →
📝 Report incorrect info about Numerai