Break-Even Analysis: The Minimum Viable Numerai Model

With the payout factor near 0.1, 50-70% of Numerai models fall below break-even each round. Here's the minimum MMC and CORJ60 your model needs to profit.

What does a Numerai model need to score before it starts making money? The break-even threshold depends on two things: raw score and payout factor. Both have shifted since the early tournament days.

The payout formula is: Payout = stake x (0.5 x CORJ60 + 2.0 x MMC) x payout_factor. CORJ60 measures your model's correlation with targets over a 60-day window. MMC (Meta-Model Contribution) measures how much unique signal your predictions add beyond what the crowd already provides. The break-even condition on score: 0.5 x CORJ60 + 2.0 x MMC > 0. A model clears this bar when its MMC contribution outweighs any drag from negative correlation. But score alone is not enough. With the payout factor near 0.1, models that break even on score earn a fraction of what the same performance returned three years ago.

How Many Models Break Even?

The break-even percentile tracks what fraction of models fall below the zero-score threshold each round.

Line chart of break-even percentile over time showing a 10-round rolling average of models below the zero-score threshold, fluctuating between 30% and 80% across rounds 200 to 1,200
Line chart of break-even percentile over time showing a 10-round rolling average of models below the zero-score threshold, fluctuating between 30% and 80% across rounds 200 to 1,200

The rolling average swings between 30% and 80%. Recent rounds sit in the 60-70% range, meaning only about a third of staked models post a positive raw score. The troughs coincide with difficult market regimes where even consensus predictions underperform.

The break-even bar is not stable. It shifts with market conditions, meta-model composition, and model crowding. The percentage below break-even has trended upward from the mid-tournament era, suggesting the field has gotten harder as participation grew.

Payout Factor Compression

Breaking even on score is necessary but not sufficient. The payout factor determines how much NMR a given score actually pays out.

Dual-axis line chart showing payout factor declining from above 0.5 to near 0.1 alongside median positive payout per round compressing from 0.3 NMR to near zero across rounds 200 to 1,200
Dual-axis line chart showing payout factor declining from above 0.5 to near 0.1 alongside median positive payout per round compressing from 0.3 NMR to near zero across rounds 200 to 1,200

The payout factor dropped from 0.5-1.0 in the early rounds to roughly 0.1 by round 500, and it has stayed compressed since. The median positive payout tracked it step for step: a model earning 0.3 NMR per round in the high-factor era earns closer to 0.03 NMR on the same score today.

Fewer models break even, and those that do earn far less per unit of stake. A participant entering today faces a higher performance bar and a lower reward for clearing it. For how the factor interacts with total tournament economics, see Round Economics.

Profitability Across Eras

Dividing tournament history into 200-round eras shows how profitability has shifted over time.

Bar chart of profitable model-rounds by era with sample sizes: 69% of 431,225 for rounds 200-400, 50% of 201,202 for rounds 400-600, 64% of 234,619 for rounds 600-800, and 53% of 327,234 for rounds 800+
Bar chart of profitable model-rounds by era with sample sizes: 69% of 431,225 for rounds 200-400, 50% of 201,202 for rounds 400-600, 64% of 234,619 for rounds 600-800, and 53% of 327,234 for rounds 800+

The early era (rounds 200-400) was the most profitable: 69% of 431,000 model-round observations posted a positive payout. The 400-600 era hit a trough at 50%, essentially a coin flip. Rounds 600-800 recovered to 64%, likely reflecting improved model quality and some weaker models exiting. The current 800+ era sits at 53% across 327,000 observations.

Profitability rates oscillate with market conditions and meta-model dynamics. But the compressed payout factor means absolute NMR earned per positive model-round is far lower than in the 200-400 era. A 53% profitability rate at a 0.1 payout factor pays less than 69% at a 0.5 factor.

The MMC Threshold

Rearranging the break-even formula gives the minimum MMC needed: MMC > -0.25 x CORJ60. When CORJ60 is negative, a model needs positive MMC to compensate. When CORJ60 is positive, the model has a buffer and can tolerate some negative MMC.

Line chart of minimum MMC needed for a positive score at the 25th, 50th, and 75th CORJ60 percentiles, showing the 25th percentile requiring 0.005-0.015 MMC while the 75th percentile can tolerate negative MMC
Line chart of minimum MMC needed for a positive score at the 25th, 50th, and 75th CORJ60 percentiles, showing the 25th percentile requiring 0.005-0.015 MMC while the 75th percentile can tolerate negative MMC

A model at the 25th percentile of CORJ60 (below-average correlation) needs roughly 0.005-0.015 MMC to break even. At the median, the threshold hovers near zero. At the 75th percentile, a model has enough CORJ60 buffer to sustain mildly negative MMC and still post a positive score.

If your model's CORJ60 is below average, MMC is your lifeline. Predictions that add unique information to the meta-model earn positive MMC. Models that are uncorrelated with the crowd and directionally correct benefit most. Models that merely copy the consensus contribute nothing unique and rely entirely on CORJ60 to stay profitable.

What a New Participant Needs Today

Here is what a minimum viable Numerai model looks like in 2026:

Beat the median on at least one axis. With 50-67% of models below break-even in recent rounds, median performance is not enough. You need above-average MMC, above-average CORJ60, or both.

Prioritize MMC if your correlation is weak. The break-even formula weights MMC four times as heavily as CORJ60 (2.0 vs 0.5). A model with poor correlation but strong meta-model contribution can still profit. The reverse — good correlation with negative MMC — faces a much narrower margin.

Size expectations to the payout factor. At a factor near 0.1, even a top-third model earns modest NMR per round. Compounding works, but over dozens of rounds, not days. The staking profitability data shows a median per-round return of 0.05% of stake. Track payout trends on the Trends page.

Expect regime dependence. In favorable regimes, 70% of models profit. In hostile ones, 70% lose. No model quality guarantees consistent profitability across all market environments. Understanding market regimes helps set expectations for drawdown periods.

The break-even bar is higher and the reward for clearing it is lower than at any prior point. But the edge remains real and persistent across rounds for models that genuinely add signal. The question is whether you can build a model that consistently lands in the top third.