A Trunkline living receipt · watch/log only · no money moved
The AI Trading Agent Scoreboard
We watch the public AI trading agents so you don’t have to. Every number below is dated, sourced, and labeled with how much we trust it, including the ones from our own bot, losses and all. No predictions, no coins to shill. Just the receipts on how “AI trading” is actually going.
688 user-configured agents across 7 LLM families on Hyperliquid/Aster over ~2 months: only 42% finished at break-even or better, with per-agent results scattered from −30% to +307%. Source: Wallet V benchmark, self-published PR via HackerNoon/Chainwire, 2026-06-15. Not independently verified. Treat as directional.
Alpha Arena Season 1, six frontier models, $10,000 each, real autonomous trading
Crypto perps season, 2025-10-18 → 2025-11-03. Sourced from iWeaver / The China Academy coverage. Final balances where verified; UNVERIFIED clearly marked.
| Agent | Result ($10k start) | Note | Confidence |
|---|---|---|---|
| Qwen 3 Max | +22.9% ($12,287) | Winner, most disciplined risk control | medium |
| DeepSeek V3.1 | ≈ +4.5% ($10,450) | Peaked ~+125% mid-competition, gave nearly all of it back | low |
| Gemini 2.5 Pro | −45.4% ($5,464) | Large drawdown | medium |
| GPT-5 | −53.3% ($4,671) | Overleveraged, weak risk control | medium |
| Claude Sonnet 4.5 | major drawdown | exact final unverified | unverified |
| Grok 4 | reported wiped out | exact final unverified | unverified |
Read that again: 2 of the 6 smartest AI models on earth made money; the rest lost up to half the account. The winner won on risk discipline, not on being smarter.
Season 1.5, the stocks-token season
US-equity tokens (TSLA/NVDA/MSFT/AMZN/NDX) on Hyperliquid, 2025-11-19 → 2025-12-03. Source: gncrypto.news / onedayadvisor / panewslab.
| Agent | Result | Note | Confidence |
|---|---|---|---|
| Mystery Model “Grok 4.20” | +11.8% ($11,178) | Winner; 10x leveraged longs (PLTR/MSFT); minor source discrepancy (+12.11% also reported) | medium |
| GPT-5.1 | 2nd | final unverified | unverified |
| Gemini 3 | 3rd | final unverified | unverified |
| Kimi K2 | N/A | new entrant; final unverified | unverified |
Our own bot, held to the same standard
We run a paper-trading bot (fake money, real signals) and log it in the same ledger we use to grade everyone else. As of the 2026-07-13 check-in: 15 trades, 20% win rate, net −$13 (paper), against an AI-agent cohort whose worst public result was −53%. Our bot was long-only in that sample and it bled in a downtrend. That finding is exactly why the system now tests short and flat variants, on paper, behind the same gates.
Why publish a losing number? Because this page is worthless if we only grade other people. Nothing we run goes live, and nothing gets sold, until it passes the gates: 100+ trades, 3+ market regimes, out-of-sample, after costs. It hasn’t yet. That’s the receipt.
The scam file RED FLAG
“Neyro Quantum Alpha,” an “AI quantum trading agent” claiming ~27.6% per month (self-reported, not on-chain verified), closed beta, 5,000 USDT minimum entry.
What the record actually shows, each item dated: fraud/Ponzi warnings from Nigeria’s SEC (Jan 2026) and the Bank of Russia (Jan 2026) against parent Aurum Foundation; New Zealand FMA warning (May 2026); France blacklist (2026-06-12); Hong Kong SFC alert list #3472 (2026-06-22); Belgium FSMA MiCA warning. A June 2026 exposé (DisruptionBanking, 2026-06-29) reported recruiters pitching it as “referral marketing” while repeating the ~30%/month claim. The project touts a phase-1 code audit as legitimacy. An audit checks code, not returns.
The tell, in plain English: anyone who has to recruit you into a guaranteed-sounding monthly return is selling you a story, not a system. Six regulators on three continents have said so with dates.
What this scoreboard actually says
The institutions ARE spending billions on AI trading, and the public evidence so far says most AI agents you can actually buy access to lose money or scatter wildly, the rare winners win on risk discipline, and the loudest “AI profits” claims come with regulator warnings attached. You can’t outspend that arms race. You can refuse to be its exit liquidity: demand dated, sourced, loss-included receipts, from every bot, every guru, and from us.
Updated 2026-07-25 · sources: iWeaver, The China Academy, gncrypto.news, panewslab, walletv.io (Chainwire PR), coincodex, SEC Nigeria, Bank of Russia, NZ FMA, HK SFC, Belgium FSMA, DisruptionBanking · confidence labels are ours. Watch/log only. This page moves no money.
Educational only, not financial advice. Crypto involves risk; you can lose money. Some results above are self-reported by third parties and marked accordingly.
Own your edge, off the algorithm
Get the free Trunkline Discipline Checklist, and deadline alerts as the dates land. No hype, one click out.
Send me the checklistFree. Educational only, not financial advice. Check your spam folder for the confirmation email.
