Verified · updated daily · as of Oct 28, 2026

Verified AI trading leaderboard

Which AI trading agents, bots and vaults actually perform? We compute every number from public chain and venue data. No screenshots, no self-reported PnL, no backtests. Security is a separate gate.

Of the 157 registered AI agents betting on prediction markets with enough activity in the last 90 days, 141 made money before AI-tool costs; across all 321 active agents, trading ROI was +9.1%. Of the 6 Hyperliquid vaults whose leaders say they're run by AI ($100k+, 30+ days live), 2 beat BTC over the same dates, 1 lost money and 1 fell more than 30% from a peak. Users of 8 AI trading apps on Hyperliquid traded $24.2M in 30 days; their realized PnL on those trades was $13k.

157registered AI agents ranked (of 321 active)
+9.1%trading ROI across all 321 active agents (90 days)
6Hyperliquid vaults that call themselves AI
1of those fell more than 30%
Evidence:AVerifiedBVenue-reportedCClaimedOnly grade A is ranked. Claims without evidence: claim check.

AI agents with a public track record

AVerified Olas Predict trader agents: autonomous AI agents registered on-chain in the Olas service registry (3,875 registered so far), each betting on prediction markets from its own Safe wallet. Every bet and payout is on-chain; we compute the results from Olas's public indexers of that data. Last 90 days, as of Oct 28, 2026. Ranked: 20+ settled bets, $25+ settled and active in the last 30 days.

Omen (Gnosis)

277agents active
138enough activity to rank
91%of those made money
+8.7%median ROI
$5,140profit on $55,481 settled (all active)
0.182median Brier

Polymarket

44agents active
19enough activity to rank
84%of those made money
+7.3%median ROI
$102profit on $2,089 settled (all active)
0.184median Brier

Read these numbers right

Profit is trading profit on settled markets: payouts minus stakes and market fees. It leaves out what each agent pays for AI predictions and any staking rewards. Stakes are small, often a few dollars a bet, and these markets are thin, so returns don't scale to big money. Agents that stopped trading drop out of the ranking, which flatters the share that made money; the fleet ROI across every active agent is the fairer number. Brier score measures how good the agent's probabilities were (0 is perfect, 0.25 is a coin flip on a 50/50 market).

Run an AI trading agent?

Register it on-chain or publish its trading address on your official site, docs or X account, then send us the name, the official page and the address. We rate it with the same method as everything here. Listing is free, and nobody can pay for a rank.

Open data: Olas Predict agent ratings (JSON)

AI trading apps on Hyperliquid

AVerified Apps that run AI agents or an AI copilot for their users tag every order with a builder code, and Hyperliquid publishes every fill that carried one. From Sep 28, 2026 to Oct 27, 2026: 8 apps, 1,561 users, $24.2M traded, and users' realized PnL of $13k on those fills. Sorted by volume; not a ranking of skill.

1
TrueNorth
283 users · 36,662 fills · no app fee on these fills

How it tradesnot clearly stated on its site

— daily volume
Volume$11.4Mlast 30 days
Users' realized PnL−$1,36151% of users in profit
Median user$0.02realized PnL
2
Senpi
831 users · 32,338 fills · $2,522 in app fees

How it tradesautonomous strategies and manual or copy trades

— daily volume
Volume$5.9Mlast 30 days
Users' realized PnL−$4,71440% of users in profit
Median user−$0.38realized PnL
3
DXAP
233 users · 15,999 fills · $894 in app fees

How it tradesautonomous agent, steered by chat

— daily volume
Volume$3.6Mlast 30 days
Users' realized PnL$24k34% of users in profit
Median user−$1.28realized PnL
4
Wayfinder
43 users · 670 fills · $644 in app fees

How it tradesagent trades on request; automated strategies

— daily volume
Volume$1.4Mlast 30 days
Users' realized PnL−$1,08042% of users in profit
Median user−$0.04realized PnL
5
Midas
80 users · 999 fills · $540 in app fees

How it tradescopilot: AI suggests, the user taps to trade

— daily volume
Volume$1.1Mlast 30 days
Users' realized PnL−$4,06759% of users in profit
Median user$0.38realized PnL
6
Katoshi
16 users · 1,405 fills · $156 in app fees

How it tradesbots from alerts, webhooks or AI agents, plus a manual terminal

— daily volume
Volume$632klast 30 days
Users' realized PnL$63238% of users in profit
Median user−$1.56realized PnL
7
Minara AI
73 users · 2,572 fills · $47 in app fees

How it tradesautopilot, AI copilot and manual orders

— daily volume
Volume$254klast 30 days
Users' realized PnL−$23747% of users in profit
Median user−$0.21realized PnL
8
Starchild
2 users · 100 fills · $1 in app fees

How it tradescopilot: orders from chat instructions

— daily volume
Volume$4,327last 30 days
Users' realized PnL−$10% of users in profit
Median user−$0.85realized PnL

Read these numbers right

Only fills that carried the app's builder code are counted, so trades sent another way (or on plans that don't add the code) are missed. Realized PnL is from closing fills and is before exchange fees; open positions aren't included. Using an AI app doesn't mean every trade was made by the AI: some apps mix autopilot, copilot and manual orders. Builder addresses come from each app's own docs or code, or DeFiLlama's open-source adapters; apps are listed only when their own site says they're AI-driven.

Open data: AI trading apps on Hyperliquid (JSON)

AI-labelled Hyperliquid vaults

AVerified The vault's name or the leader's own description on Hyperliquid says it's run by AI. That's the leader's claim, quoted on each card; we don't verify how a vault trades. The numbers are ours. As of Oct 28, 2026.

1
FuturAI Labs - High Sharpe, Low Volatility
Hyperliquid vault · since Mar 4, 2026 · $0.3M TVL

AI labelfrom the vault's own name (“AI”)

— return · - - BTC
Return+7.9%−3.6 pts vs BTCBTC +11.5%
Max drawdown−0.6%
Sortino13.9730 days +1.7%
2
Stratwise Multi-Asset Public
Hyperliquid vault · since Jul 22, 2026 · $0.4M TVL

Leader saysАutomated AI strategy trading multiple crypto assets in one vault. Volatility & squeeze forecasting via a dynam…

— return · - - BTC
Return+6.9%−15.7 pts vs BTCBTC +22.5%
Max drawdown−2.5%
Sortino5.8530 days +0.4%
3
Akka Hyper AI
Hyperliquid vault · since Nov 12, 2025 · $0.1M TVL

AI labelfrom the vault's own name (“AI”)

— return · - - BTC
Return+9.8%+30.2 pts vs BTCBTC −20.3%
Max drawdown−5.0%
Sortino2.3330 days −0.0%
4
AceVault Hyper01
Hyperliquid vault · since Oct 29, 2025 · $0.9M TVL

Leader saysAdvanced AI Quantitative Long/Short Strategy: Capitalize on high-quality asset appreciation while str…

— return · - - BTC
Return+17.1%+43.5 pts vs BTCBTC −26.5%
Max drawdown−24.2%
Sortino1.0630 days +3.7%
5
AIQuantPulse
Hyperliquid vault · since Mar 25, 2026 · $1.1M TVL

AI labelfrom the vault's own name (“AI”)

— return · - - BTC
Return+0.3%−13.2 pts vs BTCBTC +13.6%
Max drawdown−10.8%
Sortino0.1330 days −2.0%
✕
DailyTradeAI Low risk
Hyperliquid vault · since Nov 5, 2025 · $0.1M TVL · †

AI labelfrom the vault's own name (“AI”)

— return · - - BTC
Return−67.3%−45.2 pts vs BTCBTC −22.1%
Max drawdown−90.3%
Sortino0.2330 days −19.6%

Open data: AI-labelled vault ratings (JSON)

Biggest Hyperliquid vaults, any strategy

AVerified All 8 that passed both gates, out of 17 open user-led vaults with $1M+ and 90 days live. As of Oct 28, 2026.

1
DOUBLETOP Vault
Hyperliquid vault · since Jun 24, 2026 · $6.7M TVL
— return · - - BTC
Return+26.5%−6.1 pts vs BTCBTC +32.6%
Max drawdown−10.2%
Sortino6.5930 days −6.4%
2
Quantitative Market Neutral Strategy
Hyperliquid vault · since May 6, 2026 · $1.0M TVL
— return · - - BTC
Return+7.7%+8.3 pts vs BTCBTC −0.5%
Max drawdown−2.9%
Sortino3.3830 days +0.5%
3
Growi HF
Hyperliquid vault · since Oct 29, 2025 · $13.3M TVL
— return · - - BTC
Return+26.4%+52.9 pts vs BTCBTC −26.5%
Max drawdown−12.2%
Sortino1.7130 days +0.5%
4
Ultron
Hyperliquid vault · since Dec 3, 2025 · $1.1M TVL
— return · - - BTC
Return+8.7%+22.0 pts vs BTCBTC −13.3%
Max drawdown−28.8%
Sortino0.5930 days −4.4%
5
Black Ops
Hyperliquid vault · since Jan 7, 2026 · $1.4M TVL
— return · - - BTC
Return+3.4%+14.8 pts vs BTCBTC −11.4%
Max drawdown−25.9%
Sortino0.4330 days −4.9%
6
AIQuantPulse
Hyperliquid vault · since Mar 25, 2026 · $1.1M TVL
— return · - - BTC
Return+0.3%−13.2 pts vs BTCBTC +13.5%
Max drawdown−10.8%
Sortino0.1330 days −2.0%
7
Orbit Value Strategies
Hyperliquid vault · since Dec 17, 2025 · $1.4M TVL
— return · - - BTC
Return−2.5%+3.6 pts vs BTCBTC −6.1%
Max drawdown−20.6%
Sortino-0.0330 days +4.9%
8
FC Genesis - Quantum
Hyperliquid vault · since Oct 29, 2025 · $2.2M TVL
— return · - - BTC
Return−17.0%+9.5 pts vs BTCBTC −26.5%
Max drawdown−17.4%
Sortino-1.9230 days −7.1%

All 17 vault ratings, including the 9 that fell more than 30% →

Which AI model forecasts and trades best?

Public benchmarks run by others, labelled by type. Real money beats paper, and paper beats replaying the past. We link every source and don't change their numbers.

Nof1 Alpha Arena Real money

Season 1: crypto perps on Hyperliquid, $10,000 per model, Oct 18 – Nov 3, 2025.

ModelFinal return
Qwen3 Max+22.3%
DeepSeek Chat V3.1+4.9%
Claude Sonnet 4.5−42.0% to −30.8%
Grok 4−57.9% to −45.3%
Gemini 2.5 Pro−56.7% to −45.5%
GPT-5−62.7% to −58.7%

Reported final returns for the four losing models differ between sources (one may be an earlier snapshot); Nof1's own board could not be loaded to settle it. Sources: 1, 2.

Season 1.5 (US stocks): Grok 4.20 (entered as the Mystery Model) won with +12.1% overall. Seven of the eight models finished with a loss; Nof1 did not publish per-model totals in a form we could load. Source. Season 2: Not launched as of Oct 2026. Nof1 raised $15M in May 2026 and said it plans to run Season 2.

ForecastBench Live forecasting

Questions that resolve in the future, so nothing can be memorized. Baseline board, as of Oct 28, 2026. Higher Brier Index is better; human reference rows in italics.

ModelBrier IndexBrier
Superforecaster median forecast
ForecastBench		67.50.106
Public median forecast
ForecastBench		62.40.141
o3-2025-04-16-scratchpad
OpenAI		61.50.148
kimi-k3-max-128k
Moonshot AI		61.00.152
minimax-m3-adaptive-thinking-12000
MiniMax		60.90.153
claude-opus-4-1-20250805
Anthropic		60.60.156
claude-3-7-sonnet-20250219-scratchpad
Anthropic		60.60.155
o4-mini-2025-04-16-scratchpad
OpenAI		60.60.155
claude-sonnet-4-6-adaptive-thinking-16000
Anthropic		60.60.155
gpt-4.5-preview-2025-02-27
OpenAI		60.60.155
o4-mini-2025-04-16
OpenAI		60.50.156
grok-4.20-0309-reasoning
xAI		60.50.156

Data: ForecastBench, Forecasting Research Institute, CC BY-SA 4.0.

Prophet Arena Live forecasting

AI agents forecast real Kalshi and Polymarket events; ranked by Prophet Arena's skill score against the market, scored Oct 28, 2026. Skill above 0 beat the market's own prices; lower Brier is better; few resolved events means a noisy score.

#AgentSkillBrierEvents
1Agent GPT-6.1 Sol Pro (OpenCode)+0.0040.11117
2Agent Muse Wonder (OpenCode)+0.0030.12434
3Agent GPT-6 Astra (OpenCode)+0.0480.097295
4Agent Gemini 4 Argon (OpenCode)−0.0020.12934
5Agent Claude Opus 5.5 (OpenCode)+0.0420.103296
6Claude Opus 5.5+0.0350.110444
7Agent Gemini 3.8 Flash (OpenCode)+0.0350.114316
8Gemini 4 Argon+0.0300.12164
9Gemini 3.8 Flash+0.0280.117444
10Agent DeepSeek V4.1 Flash (OpenCode)+0.0270.119344

Data: Prophet Arena (UChicago SIGMA Lab).

Metaculus FutureEval (AI forecasting tournament) Live forecasting

Spring 2026 (Jan 7 – Apr 15, 2026): 173 bots, 297 questions.

  • Professional forecasters still beat the best bot team, but the gap was not statistically significant (head-to-head −1.25, 95% CI −4.87 to 2.37).
  • Among Metaculus's own reference bots, GPT-5.1 (high reasoning) with AskNews search scored best.

Source

Trading robots with a broker-read record

BVenue-reported MQL5 Signals on real accounts (its rules don't allow demo, contest or cent accounts), where at least 90% of deals were made by a trading robot (MQL5's "Algo trading" share). MQL5's signal server reads the trades from the broker; we don't compute these. Shown, not ranked; sorted by max drawdown. Numbers as shown on MQL5 on Oct 28, 2026, checked by hand.

SignalGrowthMax drawdownRobot shareSubscribers
Gold Spike
RoboForex-ECN · MT4 · since Dec 17, 2025		+271%−8.1%100%23
Gold Reaper New V2 2
CapitalPointTrading-MT5-4 · MT5 · since Oct 22, 2024
MQL5 warns: Share of days for 80% of growth is too low
MQL5 warns: Repeated no-trading-activity notices		+326%−16.9%99%61
GoldWave signal
ICMarketsSC-MT5-4 · MT5 · since Oct 21, 2025		+789%−17.5%99%41
Combo Profile 2026
VantageMarkets-Live 14 · MT5 · since Mar 3, 2026		+774%−19.3%100%31
Precise Pair Trading Pro
ICMarketsEU-MT5-5 · MT5 · since Sep 1, 2025
MQL5 warns: No trading activity in the last 10 days
MQL5 warns: Max deposit load 98.93%		+605%−23.8%98%18
UpFuji MT4
RoboForex-Pro-5 · MT4 · since May 8, 2025		+315%−27.1%99%10
Kenni Trades Gold Breakout
Fyntura-Live · MT4 · since Apr 28, 2025		+3,767%−27.9%100%21
Soma PureGold Indexes and Oil
ICMarketsSC-MT5 · MT5 · since Nov 7, 2024
MQL5 warns: 80% of growth achieved within 14 days		+863%−28.6%99%7
MSC Gold Stable Pro
NeotechFinancialServices-Live · MT5 · since May 26, 2024		+1,536%−33.7%97%31
Onyx MT5
VantageMarkets-Live 4 · MT5 · since Apr 20, 2026
MQL5 warns: Trading style has changed; part of the history is not in the statistics		+809%−36.6%100%2
MSC SuperGold Pro
NeotechFinancialServices-Live · MT5 · since Apr 23, 2025
MQL5 warns: Too much growth in the last month indicates a high risk		+16,916%−53.0%96%16

Growth on small accounts compounds fast and big growth often comes from a few days; MQL5's own warnings are shown where it raises them. Copying a signal isn't the same as having its result.

Why 3Commas, Cryptohopper and Telegram bots aren't ranked

Their results live inside each user's own exchange account, so there's no public record to check. Marketplace returns are self-reported or backtested: CClaimed at best. See what AI trading products claim and what backs it up in our claim check. For those bots, check what we can verify: how much each one is used, what each bot can do with your money, the official links and known fakes, and the incidents. Bot makers who want verified results here can contact us about read-only verification.

How we verify

Public data only

Account history from public chain and venue data: Hyperliquid's API and builder fills, and the on-chain Olas registry and prediction markets through Olas's public indexers. Nothing an operator sends us counts.

Our own math

Returns net of deposits and withdrawals, max drawdown, Sortino, and BTC over the same dates. Before the leader's profit share.

Gates

Vaults: the venue must score 50%+ on our Safety score (Hyperliquid: 50%), and a fall of more than 30% from a peak means shown, not ranked. Agents: enough settled bets and recent activity.

Full method and limits
  • Return: each step's profit over the capital at the start plus half the net deposits in that step (Modified Dietz), chained.
  • Window: vaults, the last 12 months or since the account first held $10,000 if that's later; agents, the last 90 days.
  • Agents: ROI = trading profit on settled markets over money settled; Brier = mean Brier score of settled bets; ranked by ROI, ties to more money settled. Read from Olas's public Predict indexers (Omen on Gnosis, Polymarket), which anyone can rebuild from the chain.
  • AI apps: every fill Hyperliquid publishes for the app's builder code over 30 days: volume, distinct users, and closed PnL summed per user, before exchange fees. Sorted by volume, not ranked on skill.
  • Robot signals (grade B): MQL5's own figures for real accounts with 90%+ robot trading, copied by hand with the date; shown, never ranked with grade A.
  • AI models: other people's public benchmarks, linked and unchanged. Finished seasons are entered by hand; where sources disagree we show the range.
  • Rank: Sortino ratio (return per unit of downside volatility), ties to the smaller drawdown.
  • AI label: a name that says AI, or a leader description that claims AI runs the vault ("AI strategy", "automated AI", a model name). We quote it and don't verify how the vault trades.
  • Security: a vault leader keeps at least 5% of the vault and takes a 10% profit share; depositors wait 1 day to withdraw. Hyperliquid's terms restrict US and Ontario (Canada) users and sanctioned countries.
  • Limits: provisional data (older history comes in points several days apart, so older drawdowns may be understated); vaults that closed before we started recording aren't here; past results don't predict future results.

No affiliate links on this page, and nobody can pay for a place on it. Not investment advice. Found an error? Tell us with the address. Agents can read this board through our MCP server or the open data.

Get the weekly agent risk update

New Agent Risk Scores, incidents and safety changes for the wallets, exchanges and brokers AI agents use. One email a week, unsubscribe anytime.

No spam. Unsubscribe anytime. Privacy