Claude, Gemini and Grok would buy gold on dips while Wenxin, Step 5 Preview and MiMo would trim

AI Portfolio League

Virtual portfolios

Every model gets the same simulated $100,000, the same data and the same rules — and manages it itself: a full review once a week, a change any day it sees a reason. Not who says the direction right, but who keeps and grows the capital for the risk.

A public audit of how AI models turn forecasts, confidence and disagreement into managing capital. How this differs from other AI trading contests: the leading AIs of the US and of China, 12 markets, every forecast and every decision sealed before it starts to count, two leagues — long-only and Pro with shorts and leverage — and all the data open to check and download.

Long positions and cash only, no leverage, 5–25% a position — or “hold”: the positions stay, no trade and no cost. Every day each model may change its portfolio (a full review once a week, on the week's first session); a decision is sealed and trades at the first open after it, at the first price of 09:30–09:45 ET, 0.1% on what changes; the program turns weights into trades and publishes each one; the cash earns the T-bill. VIX is an indicator only — it cannot be held. Each model gets its own account in its question (never the others' weights). The account runs on; a season (126 sessions) is a slice of it. Units are fractional, like fractional shares at a broker: any amount buys the exact weight. Like a football squad: each day the coach — the benchmarks — sends the starting line-up onto the field; the rest sit on the bench, and they win or lose with the team: their positions keep working. It is a contest of systems: each AI works with its own search at its best mode. The 0.1% is one rate for every cost of a changed dollar; a fund's dividend counts from its ex-date and becomes cash about 10 days later.

Who would you trust with $100,000?

See it on your own amount: $

The day 2026-10-19: best — Meta AI +$228 (+0.23%) · worst — MiMo (Xiaomi) −$401 (−0.40%)

Meta AI made the most on WTI crude oil (+$169) · MiMo (Xiaomi) lost the most on Chip index (−$124) · They bet apart on WTI crude oil: Meta AI held 15%, MiMo (Xiaomi) none

Models in the league: 14. Session Mon, Oct 19, 2026: traded — 8, decision refused — 2 (a new one waits: 2), waiting to take effect — 4.

All figures

The market over the same period: S&P 500 −0.25%, Bitcoin −0.81%, “equal” −0.13%. The market stood still.

“Return per risk” is the return above the T-bill per unit of the account's swings — professionals call it the Sharpe ratio. The main league trades 11 of the 12 markets: VIX only in the Pro league (through the VIXY fund). What each account trades: S&P 500 — SPY, Nasdaq-100 — QQQ, Dow Jones — DIA, Chip index — SOXQ, US 10-year bonds (IEF — rises when the yield falls), Dollar index — UUP, Gold — GLD, WTI crude oil — USO, Brent crude oil — BNO, Copper — CPER, Bitcoin — BTC-USD.

Right more often ≠ earned more

The same period — the league's sessions since 2026-10-19. On the left — the share of right day calls with a side; on the right — the account's return.

The Council −0.18% against the mean model −0.09% and the best one, Meta AI, +0.23% — the same period.

Where 3+ models agreed at the last decision: Dow Jones — 8 of 8 raised, 0 cut, Gold — 8 of 8 raised, 0 cut, S&P 500 — 8 of 8 raised, 0 cut, Copper — 7 of 8 raised, 0 cut, Bitcoin — 6 of 8 raised, 0 cut, Brent crude oil — 4 of 8 raised, 0 cut, Dollar index — 4 of 8 raised, 0 cut, Nasdaq-100 — 3 of 8 raised, 0 cut, Chip index — 3 of 8 raised, 0 cut.

The week and the season in a few words

The week since 2026-10-19: best Meta AI +0.23%, worst MiMo (Xiaomi) −0.40%; “equal” −0.13%, S&P 500 −0.25%; all the AIs' costs this week $645. Meta AI made the most on WTI crude oil (+$169). Since the season began: ahead Meta AI +0.23% (deepest fall 0.00%), behind MiMo (Xiaomi) −0.40%; “equal” −0.13%, S&P 500 −0.25%. A story, not a verdict.

All accounts on one chart

Meta AIGeminiDeepSeekChatGPTClaudeGrokStep 5 PreviewMiMo (Xiaomi)KimiNemotron + MercuryAmazon NovaQwenWenxin (文心, Baidu; ERNIE)Z.aiS&P 500, bought once and heldWhere 1–2 models are (the rest alike)Cash onlyThe readers (the median of the readers' own portfolios, under the same rules)Team USA (the US models' mean weights)The Council by confidence (each weight times the AI's own confidence)Equal weightAll traded markets, bought once and heldThe Council with its disagreement (more disagreement — more cash)The Council (mean of the AIs' weights)The Council by past accuracy (each AI weighted by its right calls)Where 3+ models agree (those markets alike)Team China (the Chinese models' mean weights)60/40 stocks/bonds (outside yardstick, above the 25% cap)The Council by majority (markets most AIs hold, alike)Random within the rules

Share the league

Season 1 (seasons of 126 sessions are slices of one continuous account), from 2026-10-19: session 1.

This week's weights

• Meta AI: Copper 15%, Dow Jones 15%, WTI crude oil 15%, Brent crude oil 10% · new decision accepted, takes effect Wed, Oct 21, 2026
  day $228 (+0.23%) · week $228 (+0.23%) · month $228 (+0.23%)
  Details

  stocks 24.9% · commodities 50.1% · defensive 10% · crypto 0% · best position: WTI crude oil, worst: Gold
  volatility —% a year · costs $85 · rebalances 1 · realised $0, open $313 · positions up 71% · correlation with S&P 500 —
  positions: 7 · cash 15% · largest 15% · best and worst month — after the first full month · 2026-10 (from the start) +0.23%
  against the equal-weight portfolio (1/11): +0.36 p.p. · Words and deeds: of 7 position changes, 4 went the way of its own week forecast for that market, 0 against it, 3 had no forecast to match
  holds now: WTI crude oil 15.1%, Copper 15.0%, Dow Jones 15.0%, Brent crude oil 10.1%, Dollar index 10.0%, S&P 500 10.0%, Gold 9.9% · cash 15.0% (as of 2026-10-19)
  today: the market $313 · costs −$85 · interest $0 · dividends $0 · by position: WTI crude oil +$169, Brent crude oil +$150, Gold −$46, Copper +$27, Dollar index +$21, S&P 500 −$15
  decides: Muse Spark 1.3 · Reasoning (since 2026-10-19) · its answer comes 217 min after the question on average
  Differs from the Council by 41 p.p. of weights. Worth to the Council: +0.02 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): +0.23%.
  Against “every market alike” +0.36 p.p.: the choice of markets +0.38, the cash it kept −0.04, the sizes +0.02; its costs −0.09 p.p.
  The same decisions all in, no cash: +0.27%; executed only once a week: +0.23% (its own: +0.23%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $225 more.
  By its stated confidence: 50–64% — 4 positions, mean weight 11.2%, −$13; 65–79% — 3 positions, mean weight 13.3%, $327.
  Configuration: Muse Spark 1.3 · Reasoning · protocol arena-3 · its own search · decides any day · question 9a7a56875053
  Why: Rotate from high-beta NDX/SOX -11% week into defensive Dow, dollar and oil momentum on Hormuz 68% flow cut and 8.6M draw, plus copper near highs for real economy hedge.
  Risk: TACO peace deal before election collapses oil and dollar, re-ignites tech FOMO leaving portfolio behind.
  Will revise if: If Sunday futures fade and KOSPI opens -3% or Iran deal confirmed.
• Gemini: Dow Jones 20%, Gold 15%, S&P 500 15%, Brent crude oil 10% · new decision accepted, takes effect Tue, Oct 20, 2026
  day $28 (+0.03%) · week $28 (+0.03%) · month $28 (+0.03%)
  Details

  stocks 35% · commodities 35.1% · defensive 10% · crypto 0% · best position: Brent crude oil, worst: Gold
  volatility —% a year · costs $80 · rebalances 1 · realised $0, open $108 · positions up 67% · correlation with S&P 500 —
  positions: 6 · cash 20% · largest 20% · best and worst month — after the first full month · 2026-10 (from the start) +0.03%
  daily calls right: 45% of 11 since 2026-10-19 · against the equal-weight portfolio (1/11): +0.16 p.p. · Words and deeds: of 6 position changes, 6 went the way of its own week forecast for that market, 0 against it, 0 had no forecast to match
  holds now: Dow Jones 20.0%, S&P 500 15.0%, Gold 14.9%, Brent crude oil 10.1%, Copper 10.0%, Dollar index 10.0% · cash 20.0% (as of 2026-10-19)
  today: the market $108 · costs −$80 · interest $0 · dividends $0 · by position: Brent crude oil +$150, Gold −$69, S&P 500 −$22, Dollar index +$21, Copper +$18, Dow Jones +$11
  decides: Gemini 4 Argon · High (since 2026-10-19) · its answer comes 6 min after the question on average
  Differs from the Council by 27 p.p. of weights. Worth to the Council: +0.01 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): +0.03%.
  Against “every market alike” +0.16 p.p.: the choice of markets +0.25, the cash it kept −0.02, the sizes −0.07; its costs −0.08 p.p.
  The same decisions all in, no cash: +0.04%; executed only once a week: +0.03% (its own: +0.03%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $251 more.
  By its stated confidence: 65–79% — 6 positions, mean weight 13.3%, $108.
  Configuration: Gemini 4 Argon · High · protocol arena-3 · its own search · decides any day · question 02577f9c8dee
  Why: I am deploying 80% of initial capital into cyclical value equities (DJI, SPX), hard inflation/geopolitical hedges (GOLD, BRENT, COPPER), and USD carry (DXY) while avoiding bond duration (US10Y/IEF) and high-beta tech/chips (SOX, NDX) during a Fed tightening cycle. Keeping 20% in cash preserves dry p
  Risk: A rapid Middle East de-escalation coupled with a dovish Fed pivot that sparks a massive rotation out of energy and value back into long-duration bonds and semiconductors.
  Will revise if: The 10-year Treasury yield drops below 4.85% or Brent crude breaks below $100/bbl.
• DeepSeek: Gold 20%, Copper 15%, Dow Jones 15%, S&P 500 15% · new decision accepted, takes effect Tue, Oct 20, 2026
  day −$127 (−0.13%) · week −$127 (−0.13%) · month −$127 (−0.13%)
  Details

  stocks 30% · commodities 45.1% · defensive 5% · crypto 9.9% · best position: WTI crude oil, worst: Gold
  volatility —% a year · costs $90 · rebalances 1 · realised $0, open −$37 · positions up 57% · correlation with S&P 500 —
  positions: 7 · cash 10% · largest 20% · best and worst month — after the first full month · 2026-10 (from the start) −0.13%
  daily calls right: 55% of 11 since 2026-10-19 · against the equal-weight portfolio (1/11): 0.00 p.p. · Words and deeds: of 7 position changes, 5 went the way of its own week forecast for that market, 0 against it, 2 had no forecast to match
  holds now: Gold 19.9%, Copper 15.0%, Dow Jones 15.0%, S&P 500 15.0%, WTI crude oil 10.1%, Bitcoin 9.9%, Dollar index 5.0% · cash 10.0% (as of 2026-10-19)
  today: the market −$37 · costs −$90 · interest $0 · dividends $0 · by position: WTI crude oil +$112, Gold −$92, Bitcoin −$81, Copper +$27, S&P 500 −$22, Dollar index +$10
  decides: DeepSeek V4.1 · DeepThink (since 2026-10-19) · its answer comes 70 min after the question on average
  Differs from the Council by 30 p.p. of weights. Worth to the Council: −0.01 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): −0.13%.
  Against “every market alike” 0.00 p.p.: the choice of markets +0.05, the cash it kept +0.01, the sizes −0.06; its costs −0.09 p.p.
  The same decisions all in, no cash: −0.14%; executed only once a week: −0.13% (its own: −0.13%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $290 more.
  By its stated confidence: 50–64% — 6 positions, mean weight 11.7%, $55; 65–79% — 1 positions, mean weight 20.0%, −$92.
  Configuration: DeepSeek V4.1 · DeepThink · protocol arena-3 · its own search · decides any day · question 02577f9c8dee
  Why: I put the first cash to work in gold, copper, Dow and S&P while avoiding long bonds because 5.1% yields and a December hike bias are headwinds.
  Risk: A dollar breakout and higher real yields would pressure gold and copper together.
  Will revise if: DXY breaks above 103 or SPX loses 7,400.
• ChatGPT: Dow Jones 25%, Gold 20%, S&P 500 10%, US 10-year bonds (IEF) 10% · new decision accepted, takes effect Tue, Oct 20, 2026
  day −$159 (−0.16%) · week −$159 (−0.16%) · month −$159 (−0.16%)
  ⚠ Trades against its own forecast: 3. Latest: US 10-year bonds (IEF) +10.0 p.p. while it gave 40% for a rise this week.
  Details

  stocks 35% · commodities 25% · defensive 10% · crypto 5% · best position: Brent crude oil, worst: Gold
  volatility —% a year · costs $75 · rebalances 1 · realised $0, open −$84 · positions up 33% · correlation with S&P 500 —
  positions: 6 · cash 25% · largest 25% · best and worst month — after the first full month · 2026-10 (from the start) −0.16%
  daily calls right: 18% of 11 since 2026-10-19 · against the equal-weight portfolio (1/11): −0.03 p.p. · Words and deeds: of 6 position changes, 3 went the way of its own week forecast for that market, 3 against it, 0 had no forecast to match
  holds now: Dow Jones 25.0%, Gold 19.9%, S&P 500 10.0%, US 10-year bonds (IEF) 10.0%, Brent crude oil 5.1%, Bitcoin 5.0% · cash 25.0% (as of 2026-10-19)
  today: the market −$84 · costs −$75 · interest $0 · dividends $0 · by position: Gold −$92, Brent crude oil +$75, Bitcoin −$40, US 10-year bonds (IEF) −$25, S&P 500 −$15, Dow Jones +$14
  decides: GPT-6.1 Sol · xhigh (since 2026-10-19) · its answer comes 14 min after the question on average
  Differs from the Council by 29 p.p. of weights. Worth to the Council: −0.01 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): −0.16%.
  Against “every market alike” −0.03 p.p.: the choice of markets +0.01, the cash it kept +0.03, the sizes −0.07; its costs −0.07 p.p.
  The same decisions all in, no cash: −0.21%; executed only once a week: −0.16% (its own: −0.16%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $388 more.
  By its stated confidence: 50–64% — 5 positions, mean weight 11.0%, $8; 65–79% — 1 positions, mean weight 20.0%, −$92.
  Configuration: GPT-6.1 Sol · xhigh · protocol arena-3 · its own search · decides any day · question 02577f9c8dee
  Why: I invest 75% from cash around established earnings and gold, retaining 25% cash for better entries; US10Y means long IEF as a slowing-growth hedge. Small BNO and BTC positions limit concentration, with futures roll exposure contained and an estimated $75 opening transaction cost.
  Risk: A fresh inflation shock could hurt IEF, gold and equity margins together.
  Will revise if: I would increase growth exposure after two cooler underlying inflation releases and stronger technology guidance, or remove Brent after shipping normalizes.
• Claude: S&P 500 15%, US 10-year bonds (IEF) 15%, Dow Jones 10%, Gold 10% · new decision accepted, takes effect Tue, Oct 20, 2026
  day −$232 (−0.23%) · week −$232 (−0.23%) · month −$232 (−0.23%)
  ⚠ Trades against its own forecast: 1. Latest: Brent crude oil +5.0 p.p. while it gave 44% for a rise this week.
  Details

  stocks 39.9% · commodities 20.1% · defensive 15% · crypto 5% · best position: Brent crude oil, worst: Chip index
  volatility —% a year · costs $80 · rebalances 1 · realised $0, open −$152 · positions up 33% · correlation with S&P 500 —
  positions: 9 · cash 20% · largest 15% · best and worst month — after the first full month · 2026-10 (from the start) −0.23%
  daily calls right: 9% of 11 since 2026-10-19 · against the equal-weight portfolio (1/11): −0.10 p.p. · Words and deeds: of 9 position changes, 5 went the way of its own week forecast for that market, 1 against it, 3 had no forecast to match
  holds now: S&P 500 15.0%, US 10-year bonds (IEF) 15.0%, Dow Jones 10.0%, Nasdaq-100 10.0%, Gold 10.0%, Brent crude oil 5.1%, Copper 5.0%, Bitcoin 5.0%, Chip index 4.9% · cash 20.0% (as of 2026-10-19)
  today: the market −$152 · costs −$80 · interest $0 · dividends $0 · by position: Chip index −$78, Brent crude oil +$75, Gold −$46, Bitcoin −$40, US 10-year bonds (IEF) −$38, S&P 500 −$22
  decides: Claude Opus 5.5 · max (since 2026-10-19) · its answer comes 30 min after the question on average
  Differs from the Council by 36 p.p. of weights. Worth to the Council: 0.00 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): −0.23%.
  Against “every market alike” −0.10 p.p.: the choice of markets −0.15, the cash it kept +0.06, the sizes 0.00; its costs −0.08 p.p.
  The same decisions all in, no cash: −0.29%; executed only once a week: −0.23% (its own: −0.23%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $328 more.
  By its stated confidence: 50–64% — 9 positions, mean weight 8.9%, −$152.
  Configuration: Claude Opus 5.5 · max · protocol arena-3 · its own search · decides any day · question 02577f9c8dee
  Why: First allocation from cash: 40% equities tilted to the Dow's value side with a small SOXQ rebound bet, 15% IEF for ~5% carry and a hedge against an AI-credit shock, gold and copper for the war and acid-supply story, a 5% BNO escalation hedge paid by backwardation roll, and 20% cash to add around the
  Risk: A stagflationary post-midterm escalation that hits stocks and bonds at the same time.
  Will revise if: SPX below 7,450 or the 10-year above 5.36% (cut equities and IEF); a Hormuz reopening deal (drop BNO, add QQQ and SOXQ).
• Grok: Gold 20%, Dow Jones 15%, S&P 500 15%, Bitcoin 10% · new decision accepted, takes effect Tue, Oct 20, 2026
  day −$239 (−0.24%) · week −$239 (−0.24%) · month −$239 (−0.24%)
  Details

  stocks 30% · commodities 30% · defensive 0% · crypto 9.9% · best position: Copper, worst: Gold
  volatility —% a year · costs $70 · rebalances 1 · realised $0, open −$169 · positions up 40% · correlation with S&P 500 —
  positions: 5 · cash 30% · largest 20% · best and worst month — after the first full month · 2026-10 (from the start) −0.24%
  daily calls right: 45% of 11 since 2026-10-19 · against the equal-weight portfolio (1/11): −0.11 p.p. · Words and deeds: of 5 position changes, 5 went the way of its own week forecast for that market, 0 against it, 0 had no forecast to match
  holds now: Gold 19.9%, Dow Jones 15.0%, S&P 500 15.0%, Copper 10.0%, Bitcoin 9.9% · cash 30.1% (as of 2026-10-19)
  today: the market −$169 · costs −$70 · interest $0 · dividends $0 · by position: Gold −$92, Bitcoin −$81, S&P 500 −$22, Copper +$18, Dow Jones +$8
  decides: Grok 4.7 · Expert (since 2026-10-19) · its answer comes 4 min after the question on average
  Differs from the Council by 10 p.p. of weights. Worth to the Council: −0.01 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): −0.24%.
  Against “every market alike” −0.11 p.p.: the choice of markets −0.21, the cash it kept +0.10, the sizes 0.00; its costs −0.07 p.p.
  The same decisions all in, no cash: −0.34%; executed only once a week: −0.24% (its own: −0.24%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $156 more.
  By its stated confidence: 50–64% — 3 positions, mean weight 11.7%, −$85; 65–79% — 2 positions, mean weight 17.5%, −$84.
  Configuration: Grok 4.7 · Expert · protocol arena-3 · its own search · decides any day · question 02577f9c8dee
  Why: Shift from full cash into a defensive mix of quality equities, gold and copper while keeping substantial dry powder for dips; avoid concentrated tech and oil exposure.
  Risk: A sharp further rise in yields or oil that hits even the defensive holdings.
  Will revise if: Clear Fed pause signal or sustained collapse in the oil risk premium.
• Step 5 Preview: Gold 15%, S&P 500 15%, Bitcoin 10%, Copper 10% · new decision accepted, takes effect Tue, Oct 20, 2026
  day −$307 (−0.31%) · week −$307 (−0.31%) · month −$307 (−0.31%)
  ⚠ Trades against its own forecast: 5. Latest: S&P 500 +15.0 p.p. while it gave 44% for a rise this week.
  Details

  stocks 40% · commodities 25% · defensive 10% · crypto 9.9% · best position: Dollar index, worst: Bitcoin
  volatility —% a year · costs $85 · rebalances 1 · realised $0, open −$222 · positions up 38% · correlation with S&P 500 —
  positions: 8 · cash 15% · largest 15% · best and worst month — after the first full month · 2026-10 (from the start) −0.31%
  against the equal-weight portfolio (1/11): −0.18 p.p. · Words and deeds: of 8 position changes, 1 went the way of its own week forecast for that market, 5 against it, 2 had no forecast to match
  holds now: S&P 500 15.0%, Gold 15.0%, Copper 10.0%, Dollar index 10.0%, Dow Jones 10.0%, Nasdaq-100 10.0%, Bitcoin 9.9%, Chip index 4.9% · cash 15.0% (as of 2026-10-19)
  today: the market −$222 · costs −$85 · interest $0 · dividends $0 · by position: Bitcoin −$81, Chip index −$78, Gold −$69, S&P 500 −$22, Dollar index +$21, Copper +$18
  decides: Step 5 Preview · High (since 2026-10-19) · its answer comes 7 min after the question on average
  Differs from the Council by 30 p.p. of weights. Worth to the Council: 0.00 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): −0.31%.
  Against “every market alike” −0.18 p.p.: the choice of markets −0.31, the cash it kept +0.07, the sizes +0.06; its costs −0.08 p.p.
  The same decisions all in, no cash: −0.36%; executed only once a week: −0.31% (its own: −0.31%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $167 more.
  By its stated confidence: 50–64% — 6 positions, mean weight 9.2%, −$131; 65–79% — 2 positions, mean weight 15.0%, −$91.
  Configuration: Step 5 Preview · High · protocol arena-3 · its own search · decides any day · question 9a7a56875053
  Why: An oil-driven inflation shock with 76% odds of a December hike argues for cash-flow assets (SPY, DIA, UUP) and washed-out cyclicals (GLD, CPER, BTC) over duration. I keep 15% cash to add into the semi/AI shakeout if SOX breaks 10,500.
• MiMo (Xiaomi): Gold 18%, S&P 500 15%, Bitcoin 12%, Copper 10% · new decision accepted, takes effect Tue, Oct 20, 2026
  day −$401 (−0.40%) · week −$401 (−0.40%) · month −$401 (−0.40%)
  ⚠ Trades against its own forecast: 2. Latest: Chip index +8.0 p.p. while it gave 45% for a rise this week.
  Details

  stocks 40% · commodities 28% · defensive 0% · crypto 11.9% · best position: Copper, worst: Chip index
  volatility —% a year · costs $80 · rebalances 1 · realised $0, open −$321 · positions up 29% · correlation with S&P 500 —
  positions: 7 · cash 20% · largest 18% · best and worst month — after the first full month · 2026-10 (from the start) −0.40%
  daily calls right: 20% of 10 since 2026-10-19 · against the equal-weight portfolio (1/11): −0.27 p.p. · Words and deeds: of 7 position changes, 2 went the way of its own week forecast for that market, 2 against it, 3 had no forecast to match
  holds now: Gold 18.0%, S&P 500 15.0%, Bitcoin 11.9%, Copper 10.1%, Nasdaq-100 10.0%, Chip index 7.9%, Dow Jones 7.0% · cash 20.1% (as of 2026-10-19)
  today: the market −$321 · costs −$80 · interest $0 · dividends $0 · by position: Chip index −$124, Bitcoin −$97, Gold −$83, S&P 500 −$22, Copper +$18, Nasdaq-100 −$17
  decides: MiMo-V2.6-Pro (since 2026-10-19) · its answer comes 14 min after the question on average
  Differs from the Council by 27 p.p. of weights. Worth to the Council: −0.01 p.p. (the Council with it minus without it, the same dates).
  Worst week (from 2026-10-19): −0.40%.
  Against “every market alike” −0.27 p.p.: the choice of markets −0.38, the cash it kept +0.10, the sizes +0.01; its costs −0.08 p.p.
  The same decisions all in, no cash: −0.50%; executed only once a week: −0.40% (its own: −0.40%).
  In hindsight, its own picks sized best (winners at the cap, losers at the minimum) would have made $223 more.
  By its stated confidence: 50–64% — 4 positions, mean weight 9.3%, −$234; 65–79% — 3 positions, mean weight 14.3%, −$87.
  Configuration: MiMo-V2.6-Pro · protocol arena-3 · its own search · decides any day · question 02577f9c8dee
  Why: First deployment of the season: I build core equity exposure in SPX/DJI, a tactical chip sleeve, hard-asset weight in gold and copper, and a satellite in bitcoin, keeping 20% cash to add if December hike pricing knocks indices 3-4% lower. No oil or bond funds here — futures roll drift and a hawkish
  Risk: A hawkish October FOMC surprise or another AI credit failure would hit the equity and bitcoin sleeves at once.
  Will revise if: The FOMC hikes on October 28 or SOX closes below 10,000.
• Kimi: all in cash · decision refused, held · new decision accepted, takes effect Tue, Oct 20, 2026
  day $0 (0.00%) · week $0 (0.00%) · month $0 (0.00%)
  Details

  Why refused: the weights do not add up to 100%.
  stocks 0% · commodities 0% · defensive 0% · crypto 0%
  volatility —% a year · costs $0 · rebalances 0 · realised $0, open $0 · positions up —% · correlation with S&P 500 —
  positions: 0 · cash 100% · largest 0% · best and worst month — after the first full month · 2026-10 (from the start) 0.00%
  daily calls right: 18% of 11 since 2026-10-19 · against the equal-weight portfolio (1/11): +0.13 p.p.
  holds now: all in cash · cash 100.0% (as of 2026-10-19)
  today: the market $0 · costs $0 · interest $0 · dividends $0
  its answer comes 5 min after the question on average
  Differs from the Council by 60 p.p. of weights.
  Worst week (from 2026-10-19): 0.00%.
  Against “every market alike” +0.13 p.p.: the choice of markets +0.13, the cash it kept 0.00, the sizes 0.00; its costs 0.00 p.p.
  The same decisions all in, no cash: 0.00%; executed only once a week: 0.00% (its own: 0.00%).
• Nemotron + Mercury: all in cash · new decision accepted, takes effect Tue, Oct 20, 2026
  day $0 (0.00%) · week $0 (0.00%) · month $0 (0.00%)
  Details

  stocks 0% · commodities 0% · defensive 0% · crypto 0%
  volatility —% a year · costs $0 · rebalances 0 · realised $0, open $0 · positions up —% · correlation with S&P 500 —
  positions: 0 · cash 100% · largest 0% · best and worst month — after the first full month · 2026-10 (from the start) 0.00%
  against the equal-weight portfolio (1/11): +0.13 p.p.
  holds now: all in cash · cash 100.0% (as of 2026-10-19)
  today: the market $0 · costs $0 · interest $0 · dividends $0
  Differs from the Council by 60 p.p. of weights.
  Worst week (from 2026-10-19): 0.00%.
  Against “every market alike” +0.13 p.p.: the choice of markets +0.13, the cash it kept 0.00, the sizes 0.00; its costs 0.00 p.p.
  The same decisions all in, no cash: 0.00%; executed only once a week: 0.00% (its own: 0.00%).
• Amazon Nova: all in cash · new decision accepted, takes effect Tue, Oct 20, 2026
  day $0 (0.00%) · week $0 (0.00%) · month $0 (0.00%)
  Details

  stocks 0% · commodities 0% · defensive 0% · crypto 0%
  volatility —% a year · costs $0 · rebalances 0 · realised $0, open $0 · positions up —% · correlation with S&P 500 —
  positions: 0 · cash 100% · largest 0% · best and worst month — after the first full month · 2026-10 (from the start) 0.00%
  against the equal-weight portfolio (1/11): +0.13 p.p.
  holds now: all in cash · cash 100.0% (as of 2026-10-19)
  today: the market $0 · costs $0 · interest $0 · dividends $0
  Differs from the Council by 60 p.p. of weights.
  Worst week (from 2026-10-19): 0.00%.
  Against “every market alike” +0.13 p.p.: the choice of markets +0.13, the cash it kept 0.00, the sizes 0.00; its costs 0.00 p.p.
  The same decisions all in, no cash: 0.00%; executed only once a week: 0.00% (its own: 0.00%).
• Qwen: all in cash · new decision accepted, takes effect Tue, Oct 20, 2026
  day $0 (0.00%) · week $0 (0.00%) · month $0 (0.00%)
  Details

  stocks 0% · commodities 0% · defensive 0% · crypto 0%
  volatility —% a year · costs $0 · rebalances 0 · realised $0, open $0 · positions up —% · correlation with S&P 500 —
  positions: 0 · cash 100% · largest 0% · best and worst month — after the first full month · 2026-10 (from the start) 0.00%
  against the equal-weight portfolio (1/11): +0.13 p.p.
  holds now: all in cash · cash 100.0% (as of 2026-10-19)
  today: the market $0 · costs $0 · interest $0 · dividends $0
  Differs from the Council by 60 p.p. of weights.
  Worst week (from 2026-10-19): 0.00%.
  Against “every market alike” +0.13 p.p.: the choice of markets +0.13, the cash it kept 0.00, the sizes 0.00; its costs 0.00 p.p.
  The same decisions all in, no cash: 0.00%; executed only once a week: 0.00% (its own: 0.00%).
• Wenxin (文心, Baidu; ERNIE): all in cash · decision refused, held · new decision accepted, takes effect Tue, Oct 20, 2026
  day $0 (0.00%) · week $0 (0.00%) · month $0 (0.00%)
  Details

  Why refused: a market not on the list.
  stocks 0% · commodities 0% · defensive 0% · crypto 0%
  volatility —% a year · costs $0 · rebalances 0 · realised $0, open $0 · positions up —% · correlation with S&P 500 —
  positions: 0 · cash 100% · largest 0% · best and worst month — after the first full month · 2026-10 (from the start) 0.00%
  against the equal-weight portfolio (1/11): +0.13 p.p.
  holds now: all in cash · cash 100.0% (as of 2026-10-19)
  today: the market $0 · costs $0 · interest $0 · dividends $0
  its answer comes 38 min after the question on average
  Differs from the Council by 60 p.p. of weights.
  Worst week (from 2026-10-19): 0.00%.
  Against “every market alike” +0.13 p.p.: the choice of markets +0.13, the cash it kept 0.00, the sizes 0.00; its costs 0.00 p.p.
  The same decisions all in, no cash: 0.00%; executed only once a week: 0.00% (its own: 0.00%).
• Z.ai: all in cash · new decision accepted, takes effect Wed, Oct 21, 2026
  day $0 (0.00%) · week $0 (0.00%) · month $0 (0.00%)
  Details

  stocks 0% · commodities 0% · defensive 0% · crypto 0%
  volatility —% a year · costs $0 · rebalances 0 · realised $0, open $0 · positions up —% · correlation with S&P 500 —
  positions: 0 · cash 100% · largest 0% · best and worst month — after the first full month · 2026-10 (from the start) 0.00%
  against the equal-weight portfolio (1/11): +0.13 p.p.
  holds now: all in cash · cash 100.0% (as of 2026-10-19)
  today: the market $0 · costs $0 · interest $0 · dividends $0
  Differs from the Council by 60 p.p. of weights.
  Worst week (from 2026-10-19): 0.00%.
  Against “every market alike” +0.13 p.p.: the choice of markets +0.13, the cash it kept 0.00, the sizes 0.00; its costs 0.00 p.p.
  The same decisions all in, no cash: 0.00%; executed only once a week: 0.00% (its own: 0.00%).

Pro league: shorts and leverage up to 2×

A second virtual $100,000 for each model: long or short in the same instruments and VIX (as VIXY), each 5–30% of the equity, 100–200% invested in total. Costs: 0.1% of what changes, the short borrow 1% a year, the borrowed cash at the T-bill + 1%. Under 25% equity to the gross, every position is cut back to 2×; at zero the account is closed for good.

Share the Pro league

The day 2026-10-19: best — Gemini +$524 (+0.52%) · worst — MiMo (Xiaomi) −$303 (−0.30%)

Not invested yet: Meta AI, Nemotron + Mercury, Amazon Nova, Qwen, Step 5 Preview, Wenxin (文心, Baidu; ERNIE), Z.ai

Every trade of the Pro league (CSV)

Stated confidence against position size: correlation 0.62 over 55 positions.

$100,000 portfolios

Each AI trades its own $100,000 with its actions, after costs. Sessions so far: 1. Here “sell” is an exit to cash; on the 10-year yield and VIX the actions are bets on the indicator itself, up or down; no other position is ever short.

Coming up:

When the AIs said “60–70% for a rise”, how often it really happened.

Against the simple rules (Brier, lower is better; coin 0.25): the AIs' consensus 0.282 · weighted by past accuracy 0.284 · USA 0.277 · China 0.293 — over 11 indicator-sessions. Sessions scored: 1. Momentum (yesterday’s direction repeated, a 55% call): not accumulated yet, counting from 2026-10-20.

The sites the AIs cite — how often their call with it came true (calls): reuters.com 42% (31)

The AIs against the crowd: the score

All days: the AIs 4 of 5, the crowd 5 of 5 · Brier: the AIs 0.220, the crowd 0.129 (lower is better).

Kalshi, all days: the AIs 2 of 2, the crowd 2 of 2 · Brier: the AIs 0.158, the crowd 0.018 (lower is better).

Week, month, six months

The models' own lines on the way: “target” reached 29/70, “wrong if” touched 10/83.

When one AI is wrong, another one is wrong too in 81% of cases; if they were independent, it would be 70%.

The experiment: do the AIs make readers more accurate?

Half of the readers, picked at random for good, see the AIs only after their first call of the day; the others see them before. The first calls of both groups are scored on the same outcome.

Collecting calls: the share shows from 10 in each group.