Apr 28, 2026

Numerai Payout ROI by Stake Tier: Do Large Stakers Earn More?

Numerai payout ROI peaks at the 1K-10K NMR tier (8.1 bps median per round, 57.8% win rate). The 10K+ NMR tier earns just 0.9 bps. Scale helps, then the edge fades.

The 1K-10K NMR tier is the sweet spot of Numerai staking. Across resolved Classic rounds, models in that bracket earn a median 8.1 basis points of ROI per round and win 57.8% of the time. The 10K+ NMR tier falls to 0.9 bps and a 50.9% win rate.

This post groups every staked Numerai Classic submission since 2019 into five tiers by selected_stake_value, then computes per-NMR returns. The aggregate inequality picture lives in Payout Gini; this one tests whether scale buys a higher rate of return.

Median ROI Per Round Climbs, Then Drops

Returns rise from micro to large, then fade in the 10K+ tier.

Bar chart of median per-round ROI in basis points by stake tier, with micro at 5.1 bps, small at 5.9, medium at 5.7, large at 8.1, and largest 10K+ at 0.9

Micro stakes under 10 NMR earn 5.1 bps per round at the median across 3.0M model-rounds. Small (10-100 NMR) and medium (100-1K) sit in a tight band at 5.9 and 5.7 bps. Large stakers (1K-10K) jump to 8.1 bps, a 59% premium over micro. The curve then breaks: the small set of models staking 10K+ NMR median just 0.9 bps across 7,378 model-rounds.

The 10K+ bucket is smaller, so treat it cautiously, but it still spans 7,378 model-rounds across many resolved rounds. When even the median rate falls, the issue is unlikely to be only one bad week. Concentration risk is one possible explanation: a single underperforming large model can drag the cohort down because there are not enough peers to average against.

Win Rate Mirrors the ROI Curve

The fraction of rounds with positive payout follows the same shape as the ROI medians.

Bar chart of win rate by tier, with micro at 54.9%, small at 55.0%, medium at 55.3%, large at 57.8%, and largest 10K+ at 50.9%, against a 50% reference line

The first three tiers cluster around 55%, the large tier reaches 57.8%, and the 10K+ tier drops to 50.9%. Across 7,378 model-rounds, that is roughly the boundary where the largest stake tier stops reliably earning the payout factor and starts trading variance for the same expected NMR.

Win rate alone does not capture magnitudes. A 51% win rate with skewed positive tails could still earn well in aggregate. But when both win rate and median ROI compress together, the cohort underperforms on both axes. Crowding may be one explanation, but this chart does not isolate it from model mix or account-level strategy.

The Largest Tier Has Tighter Tails Than Smaller Stakers

ROI distributions overlap heavily across tiers, but 10K+ outcomes cluster more tightly around zero.

Boxplot of per-round ROI distribution by tier showing the 5th to 95th percentile range, all tiers centered near zero with similar interquartile ranges and the 10K+ tier showing notably tighter whiskers

The 25th-75th percentile ROI band is roughly -0.15% to +0.4% across all tiers. Most stakers see modest gains and losses with the median pinned near zero. The 95th percentile reaches 2.4% for medium and small tiers but only 1.5% for the 10K+ tier. Its 5th percentile is also less negative at -0.6% versus -1.1% for small stakers.

That tight envelope is consistent with a portfolio effect inside large stakers' accounts. If large stakers spread capital across multiple models, that would smooth returns toward the mean, which sits near zero given the April 2026 payout factor of about 0.10. Smaller stakers can swing harder in both directions; the largest tier mostly cannot.

Large Stakers Take Home More Net NMR Than Every Other Tier Combined

Per-NMR rates aside, the absolute NMR flow concentrates in the 1K-10K bracket.

Cumulative net NMR earned by stake tier from 2020 to 2026, with the large tier reaching about 700K NMR, medium and largest 10K+ near 300K each, small at 70K, and micro near zero

The large tier (1K-10K NMR) has accumulated roughly 695K net NMR since 2020, more than every other tier combined. Medium (317K) and 10K+ (297K) cluster as the secondary cohort. Small stakers picked up 72K, and micro stakers ended up roughly break-even at 7K NMR across 3M observations.

Two patterns stand out. First, the 10K+ curve was flat through 2021 and only began climbing in 2022, building steadily through 2026. Very large stakes are a recent feature of the tournament, not a structural element of the early years. Second, micro stakers contribute almost zero net NMR despite producing 72% of all observations. Their tiny base sizes mean a flat ROI translates into near-zero absolute earnings.

The same pattern drives the stake concentration numbers: a few hundred large-stake models move most of the tournament's NMR, while thousands of micro stakers exist mostly as participation. See Stake Flows for how individual cohorts grow and shrink over time.

What This Means for Stakers

The per-NMR edge sits in a narrow band of stake sizes:

Below 100 NMR: ROI is roughly flat at 5-6 bps. You earn the tournament average.
100-1,000 NMR: Same per-NMR return (5.7 bps), but absolute NMR starts to matter. This tier has accumulated 317K NMR net.
1,000-10,000 NMR: The premium tier. 8.1 bps median, 57.8% win rate, 695K cumulative NMR.
10,000+ NMR: Diminishing returns. Median ROI drops to 0.9 bps and win rate falls toward 50%.

The 10K+ tier is not losing money: it still produced 297K of net NMR and earns payout factor-adjusted absolute NMR earnings that micro stakes cannot realistically match. But the per-NMR edge belongs to the 1K-10K range, where models are large enough to matter without giving up as much rate of return. For round-by-round economics see Round Economics; for whether staking pays at all see Is Staking Profitable?.