Classic vs Signals vs Crypto: Comparing Numerai's Three Tournaments
A data-driven comparison of participation, stake, returns, and score difficulty across Numerai's Classic, Signals, and Crypto tournaments.
Numerai runs three concurrent tournaments on one platform. Classic (tournament 8) provides free obfuscated features. Signals (tournament 11) requires participants to bring their own financial data. Crypto (tournament 12) targets cryptocurrency tokens and is not traded by Numerai's hedge fund. This post compares all three on participation, stake, returns, and score difficulty using nmrdash data.
Each tournament feeds Numerai's meta-model differently. For background on each format, see How Numerai Works; the tournament overview tracks live aggregate stats.
Participation
Classic dwarfs the other two tournaments in staked models.

Classic climbed from near zero in 2020 to roughly 7,000 staked models by early 2025, then dropped back to around 5,000. Signals has held steady near 800–1,000 models, capped by the cost of sourcing private financial data. Crypto, the newest tournament, sits below 500 models.
Return Distributions
Per-round payouts (payout divided by stake) show how often models win or lose relative to what they put up.

All three distributions center on zero — no tournament hands out free money — but the shapes differ in instructive ways now that the histogram is normalized to density. Classic (n≈280,802 round-model observations) is the tallest and narrowest: its zero bin reaches a density above 50, meaning the overwhelming majority of rounds land within roughly ±1%. Crypto (n≈7,935) peaks around density 42 at zero and also stays tight, with almost nothing outside ±5%. Signals (n≈11,263) is the flattest of the three, peaking near density 35 but carrying visibly wider shoulders — its ±2% to ±5% bars are the tallest among the three, confirming that Signals rounds swing harder per round than Classic or Crypto. None of the three shows meaningful mass beyond ±6%. See per-round payouts for the live data.
Score Difficulty
Median MMC per round shows which environments make positive scores hardest to come by. MMC (Meta Model Contribution) measures how much a model improves Numerai's ensemble beyond what the meta-model already captures.

Classic's median MMC hugs zero across its entire history — a side effect of a large, diverse field smoothing out per-round noise. Signals runs slightly wider. Crypto is the loudest: median MMC spikes above 0.05 and below -0.02 in several rounds between round 800 and 1,050, consistent with a thinner field where a handful of models can move the median.
Where the Money Is
Total staked NMR shows the economic weight behind each tournament.

Classic holds the clear majority of staked NMR through the full history. Signals and Crypto appear as a growing wedge on top of the stack from about round 900 onward, together reaching roughly 15–20% of total stake by round 1,200. The stake gap is wider than the participation gap because Classic attracts the largest individual stakers.
Which One Should You Play?
Classic suits participants who want to focus purely on modeling. The free data and large community lower the barrier to entry. The tradeoff is signal crowding.
Signals rewards participants with access to alternative data sources. If you have unique financial data, Signals lets you monetize it without competing against everyone using the same features.
Crypto appeals to participants with cryptocurrency domain expertise. The smaller field means less competition, but predictions are not traded by Numerai's hedge fund.
Takeaways
Classic is the main event: it leads on participation, stake, and history, and most rounds on the dashboard reflect its activity. Smaller tournaments run wider score distributions, which means more opportunity — and more risk — per round. The three tournaments are not redundant; each demands different data infrastructure and modeling approaches. Diversifying across all three is possible but expensive in both data sourcing and operational overhead.