nmrdash

Numerai dashboard.

Independent Numerai analytics, from live round scores and NMR staking to model performance, participation and reported holdings. Explore the charts below or read the research notes for methods and historical context.

Live round performance

Classic · Stake-weighted Meta Model · One line per unresolved round, aligned by scoring day.

Median latest CORR0.0239

59 scored rounds · 4 awaiting scores

Above zero now59 / 59

0 below zero

Improving over 5 updates33 / 54

21 weakening · 5 need more history

One roundLatest scoreMiddle 50% of roundsMedian by scoring day

CORR60 · 59 round paths · Each path stops at its latest available score

Loading round paths…

Scores as of (UTC) · Each metric becomes final separately.

Each line follows one currently unresolved round through its observed scoring days. The median and shaded range use the rounds available at each day, with at least 5 rounds required; fewer rounds reach the later days. The range describes the spread of scores, not a confidence interval. Missing observations remain gaps. Source: Numerai’s stake-weighted Meta Model, combining predictions with at least 0.1 NMR staked.

Stake over time

Historical NMR staked in the Numerai tournaments.

Loading stake history…

Classic as of

Signals as of

Crypto as of

Source: Numerai’s overview histories. Coverage varies by tournament; gaps indicate missing observations. These historical series may differ from current stake totals.

Stake concentration

Percentage of Classic round stake held by the top 10 accounts. A rising share means stake is becoming more concentrated; a falling share means it is more broadly distributed.

58.51%Top 10 accounts · Round 1,383 · Oct 23, 2026 (UTC)

History begins Oct 12, 2019. The shaded range shows the lowest and highest possible share where some owners are unknown. The solid line shows verified values. Missing stake observations remain gaps.

Verified sharePossible range · unknown owners

Loading stake concentration…

Stake collected · Ownership collection · Round opening dates (UTC).

Source: Numerai’s Classic round stakes and public account ownership. Known accounts’ submitting models are combined before ranking each round. For unknown owners, the lower bound treats each model as a separate account; the upper bound assigns all unknown stake to the largest known account. These bounds use public ownership collected later, which may differ from ownership at the time. Mappings are cached for up to 30 days. This measures submitting models’ selected round stake.

AUM over time

Numerai GP LLC’s reported 13F holdings value, a partial view of fund AUM.

Latest reported 13F value$1.22B

Q2 2026 · Quarter ended

Loading AUM history…

Collected · Quarter-end values in nominal USD. Missing quarters remain gaps.

Source: SEC EDGAR filings for Numerai GP LLC (CIK 0001668527). Values are filing summary totals, with amendments incorporated. 13F excludes shorts, cash, most non-US securities, and confidential holdings; it is not total fund or regulatory AUM.

Classic model-to-model performance correlation

How closely the top-staked models’ 20-day CORR scores move together over 26 prior finalized weekly rounds. The shaded band contains the middle 50% of measurable model pairs.

Median pair0.476

Top 100 by round stake · Round 1,383 · Oct 23, 2026 (UTC)

Each pair needs 20 shared scores. Long gaps wait for 26 finalized weekly scores after a CORR-version change. Incomplete stake rosters also remain gaps.

Loading model-pair correlation history…

Collected · Dates identify the round selecting the top-staked models (UTC).

Source: Numerai’s historical round stakes and finalized 20-day CORR scores. Each round selects its own top 100 or 200 models. Each distinct pair needs 20 shared scores in a window of 26 weekly rounds. Measurable pairs have equal weight in the median and quartiles; missing pairs are excluded and pair-count coverage is reported. Each new CORR version needs its own 26 finalized weekly scores, creating gaps while that history accumulates. This measures performance co-movement; stock-prediction correlations are not available in this source. Historical scores may include revisions.

Model performance over time

Each line shows the median final model score for a round. Shading is one standard deviation above and below the median.

Classic · Final per-round CORR20 and CORR60 · Equal model weights.

Published model scores are backfilled across history. Hover for scored-model and roster coverage.

Loading model performance…

Collected · Dates are round opening dates (UTC). Each scoring horizon becomes final independently.

Source: Numerai’s final CORR20 and CORR60. Each round uses all available published model scores with equal weight. The band is median ±1 population standard deviation of those scores. Missing scores are excluded. Partial historical rosters are included and identified in tooltips. Gaps mark unavailable scores or score-definition changes. No rolling average is applied.

Stake-weighted model performance

Classic models’ average CORR, weighted by their stake in each round. Each line adds final round scores over time.

CORR2015.0380 cumulative CORRFrom Feb 22, 2020 · round 2001,160 final rounds · through round 1361
CORR608.0142 cumulative CORRFrom Sep 28, 2024 · round 844478 final rounds · through round 1321

Each sum starts at its own first available final score, so totals cover different periods. CORR20 and CORR60 become final separately. These sums are score indices, not compounded returns or Meta Model CORR.

Loading performance history…

Gaps mark missing final scores or changes in Numerai’s score configuration. Sums continue across configurations; comparisons across versions need care.

Collected · Latest closed round 1383

Source: Numerai’s round model scores and stake weights. Average = sum(stake × CORR) / scored stake, including all positive-stake submitting models. Missing scores are excluded; incomplete staker rosters are excluded. Changes may also reflect changing score coverage. The integration_test benchmark supplies each metric’s scoring day and finality, not its score. No adjustment to estimate Meta Model CORR is applied.

Staking return distribution

How the largest, mid-sized, and smaller stakes performed. Compare the typical resolved position with the spread of returns and total NMR payouts.

MedianMiddle 50%10th–90th percentile

Classic · Legacy continuous staking · 60 resolved rounds, R1283–R1342

Round openings: Jun 5, 2026–Aug 27, 2026 · Latest settlement: Sep 29, 2026 (UTC). Atomic rounds are still pending; this view contains legacy payouts only.

Round stake sizeResolved position returnMedianProfitableNet payout (NMR)
Largest stakesAbove the 90th percentile
Median+0.12%
Profitable64.8%
Net payout+41,918.15 NMR
Mid-sized stakes50th–90th percentiles
Median+0.14%
Profitable65.5%
Net payout+2,345.5 NMR
Smaller stakesAt or below the 50th percentile
Median+0.16%
Profitable68%
Net payout+32.51 NMR

Collected

Source: Numerai’s resolved round stakes and settled payouts. Position return = settled NMR payout ÷ that model’s stake in that round. Each model-round has equal weight in the median, percentiles, and profitable share; zero payouts count as unprofitable. Pending payouts are excluded.

Size groups are re-ranked in each round: above the 90th stake percentile, 50th–90th, and at or below the 50th. Ties stay in the lower group, so group sizes can differ from 10% / 40% / 50%. These are model-round outcomes, not unique accounts, portfolio returns, or annual yields. Legacy and atomic staking use separate samples.

Models over time

Models that submitted predictions, split by whether they have stake.

Models with stake1,423
Models without stake13,071

Classic · All history · Latest round 1,383 · Oct 23, 2026 (UTC)

Model history covers closed rounds. Missing observations remain gaps.

Loading models history…

Collected · Dates are round opening dates (UTC).

Source: Numerai’s per-round submitted and staked model counts. Each submitting model is counted once per round.

Users over time

Unique users with a submitting model, split by whether any of their submitting models has stake.

Users with stake301
Users without stake1,970

Classic · All history · Latest round 1,383 · Oct 23, 2026 (UTC)

Historical counts use identifiable accounts where ownership is incomplete. These counts may undercount users; stake groups use identified models. Hover for ownership coverage. Missing rosters remain gaps.

Loading users history…

Collected · Dates are round opening dates (UTC).

Source: Numerai’s submitting models and public account ownership. Historical counts deduplicate identifiable accounts. “With stake” means at least one identified model had stake; “without stake” means no identified model had stake. Models without a public owner are excluded, so older counts are incomplete and stake grouping may change as ownership is resolved. Ownership uses public records observed at collection and may differ from ownership during the round. Mappings are cached for up to 30 days. Incomplete model rosters are excluded.