10-year US Treasury yield: spread 40 pp. Highest Qwen (70%), lowest Amazon Nova (30%).“Treasury yields edge higher as fed rate hike expectations remain firm [].” (Qwen)
Bitcoin: spread 27 pp. Highest Amazon Nova (60%), lowest ERNIE (33%).“BTC rebounding on macro easing and ETF inflows.” (Amazon Nova)
Chip index: spread 25 pp. Highest Amazon Nova (55%), lowest Qwen (30%).“SOX plunged on NAND oversupply fears but China’s targeted stimulus and AI hardware demand may reverse trends.” (Amazon Nova)
Next event
Thu, Nov 19, 2026 8:30 AM ET — weekly jobless claims; Fri, Dec 4, 2026 8:30 AM ET — the US jobs report
Yesterday's result
Nov 13, 2026: the Council was right on 8 of 11 markets with a move. Misses: S&P 500, Dow Jones, 10-year US Treasury yield.
A live, verifiable track record, not a backtest
Forecasts are sealed before the market opens.
An external timestamp (Bitcoin, RFC 3161) that we do not control.
An open ledger: anyone can recompute every result.
Starting line, not a rating: 100 forecasts collected, the threshold for a conclusion is 350
Right now · what this is and the scores
Where the virtual accounts stand, highest first: holding cash $100,281 · a 60/40 stock/bond portfolio $98,910 · the S&P 500 (bought and held) $98,177 · the Council $96,398.
Measured on 100 comparable questions over 3 weeks; a conclusion needs 350. This is the standing, not a verdict.
The crowd is more accurate than the Council by 1.0 points (crowd 74.0%, Council 73.0%), 100 questions.
Council minus crowd, share of right sides: 95% interval from −7.9 to 5.7 points, 100 questions.
Still noise, no conclusion drawn: 100 questions so far.
The league of virtual accounts: the AIs and the crowd side by side
The league. Session Fri, Nov 13, 2026: traded — MiMo (Xiaomi), Claude, DeepSeek, Qwen, StepFun · answered and kept the weights — ChatGPT, Z.ai, Gemini, Kimi, Grok, Meta AI · on the bench, positions carried — ERNIE (Wenxin) · positions unchanged — Nemotron, Amazon Nova.
Marks: ▲ up · ▼ down · ● even (the heat map); ● traded at the session · ○ kept its weights (the league).
Every AI keeps three separate virtual $100,000 accounts, so one AI has three figures: the League (buying only), the Pro league ( and leverage up to 2×) and the portfolio by actions (from its calls on the markets).
2-minute tour: how the league works and how to check a forecast yourself (recorded Nov 5, 2026)
Panels of AIs often agree for the same reasons and err together. Here it is measured every day where the close is the ground truth: every forecast sealed before it counts, scored on four , with the agreement, the errors made together and the lines the models drew themselves.
Start here
Every day, the world's leading AIs each get a virtual $100,000 and trade 12 markets.
Since Oct 19 the Council (their joint call) is at $96,398; the S&P 500, bought and held, is at $98,177.
On the same questions the Council's side was right 73.0% of the time; the crowd (people betting real money on Polymarket and Kalshi): 74.0%.
Every forecast is sealed before the market opens, so nobody can rewrite it.
Four virtual accounts started the season together: the AI Council, the S&P 500 bought once and held, cash, and a 60/40 stock/bond portfolio. Values as of 2026-11-13.
Account
Value, virtual $
Cash
$100,281
60/40 stock/bond portfolio
$98,910
S&P 500, bought and held
$98,177
AI Council
$96,398
Right now the Council is behind the S&P 500 by $1,779.
That says little yet: 100 comparable questions are counted of the 350 needed for a statistical verdict.
All money here is virtual. A few sessions say little about skill.
The AIs against the crowd
Start with virtual $100,000 of play money: nothing is paid or charged. Spread it over the same traded markets as the Council (0 to 25% each; the rest stays cash). Your picks stay in this browser only and are sent nowhere.
This browser blocks local storage, so your picks cannot be kept.
Humans vs machines
Seal a forecast before the result: pick a market and your chance that it closes higher at the next session than at its close of 2026-11-13. The seal is a SHA-256 hash made in your browser, the same kind the Council's seals use; anyone can recompute it from the text.
Shared table
Council or crowd: who wins the week?
A bet with play money. The week goes to whoever has the lower Brier score on the questions settled during that week: the Council, or the crowd of the prediction markets. Lower wins.
Stake $1,000, even odds.
Play with the rules
All of this runs in your browser. Nothing is sent anywhere, no model is called, and the money is virtual.
Prompter: change a belief with a fact
Rule: the odds are multiplied by the strength factor (1.2, 2 or 4) raised to the source weight (0.3, 0.6 or 1). The result stays between 1% and 99%.
Hash duel
Battle of the titans: your rule against closed questions
One rule per line: if ai > crowd + 10 then yes 5. Fields: ai, crowd, gap (ai minus crowd); signs > < >= <=; then yes or no and a stake in virtual dollars. The first matching line bets.
Virtual dollars on past questions; a backtest of your own rule, not a forecast.
Prompt drift: how much the answer moves with the wording
BRENT: spread 0.58 points across 3 wordings (from 50% to 51%).
BTC: spread 0.58 points across 3 wordings (from 49% to 50%).
COPPER: spread 1 points across 3 wordings (from 50% to 52%).
A wide spread means the answer depends on the wording, so it should be trusted less.
Table: the AIs and the market baseline
Scale 0–100: 50 = same as always saying "50%", 100 = perfect. The market is the first row, scored the same way, for comparison. A place needs at least 10 closed questions.
Out of 9 events, the AI called the move before the price in 0.
Mirror of fear
How far the AIs disagree about the fear index (VIX) for the next day: spread 4.8 points, average chance of a rise 56%. A wide spread means the AIs are unsure too.
An index of disagreement, not a forecast of the VIX.
Council ghost: earlier model versions
ChatGPT: GPT-6.1 Sol (from 2026-10-18) gave 22 answers, then GPT-6 (from 2026-10-18) gave 6 answers.
Gemini: Gemini 4 Argon (from 2026-10-18) gave 26 answers, then Gemini 3.8 Flash (from 2026-10-22) gave 1 answers.
Qwen: Qwen3.7-Plus (from 2026-10-20) gave 1 answers, then Qwen3.8-Max (from 2026-10-25) gave 21 answers.
ERNIE (Wenxin): ERNIE (from 2026-10-19) gave 3 answers, then 文心 5.1 (from 2026-10-31) gave 13 answers.
ERNIE (Wenxin): 文心 5.1 (from 2026-10-31) gave 13 answers, then ERNIE 5.1 (from 2026-11-15) gave 1 answers.
Z.ai: GLM-5.3 (from 2026-10-19) gave 11 answers, then GLM-5.3-Flash (from 2026-10-22) gave 14 answers.
Answers of earlier model versions asked in parallel on the same questions: 9.
Only answers that were really given are shown; no earlier version is invented.
Is a newer model version more accurate?
ChatGPT: GPT-6.1 Sol was right on the side in 56% of 126 scored questions, GPT-6 in 61% of 44: higher.
Gemini: Gemini 3.8 Flash was right on the side in 50% of 10 scored questions, Gemini 4 Argon in 52% of 182: about the same.
Z.ai: GLM-5.3-Flash was right on the side in 62% of 175 scored questions, GLM-5.3 in 56% of 36: lower.
A gap under 5 points counts as the same. Samples are small, and the questions and the market change between periods, so this is not proof either way.
Anti-herd points
Gemini: 3 points, changed its side on 3 of 12 markets since its previous answer.
Grok: 2 points, changed its side on 2 of 11 markets since its previous answer.
DeepSeek: 2 points, changed its side on 2 of 12 markets since its previous answer.
A point is given for changing the side (above or below 50%) with a stated reason. It counts the change only, not whether the new side was right.
Museum of the seal
Morning forecast
After the 9:30 open
After the close
2026-11-09
09:22 ET · 1091 lines
10:00 ET · 1096 lines
17:00 ET · 1110 lines
2026-11-10
09:24 ET · 1166 lines
09:31 ET · 1166 lines
16:34 ET · 1178 lines
2026-11-11
09:25 ET · 1214 lines
10:00 ET · 1215 lines
17:00 ET · 1222 lines
2026-11-12
09:29 ET · 1259 lines
09:34 ET · 1260 lines
18:00 ET · 1270 lines
2026-11-13
09:25 ET · 1303 lines
09:31 ET · 1304 lines
16:36 ET · 1319 lines
Each cell is a sealed list of the books: the last one before the open, the first one after the open and the first one after the close; lines are the lines of the Council book it holds.
Herd index
79%
How alike the AIs' probabilities are today (100% = they all say the same).
When all the AIs agree too much, that is a reason to doubt.
Where the AIs disagreed most
10-year US Treasury yield: 30% (Amazon Nova) vs 70% (Qwen)
Bitcoin: 33% (Wenxin (文心, Baidu; ERNIE)) vs 60% (Amazon Nova)
Chip index: 30% (grok) vs 55% (Amazon Nova)
Hunt the herd: 81 bets, 0 right, leader Qwen
30 settled bets collected: the table may be read, still as an observation.
AI
Bets
Settled
Right hunts
Points
DeepSeek
5
5
0
-5
grok
5
5
0
-5
Qwen
5
5
0
-5
ChatGPT
6
6
0
-6
Claude
6
6
0
-6
Gemini
6
6
0
-6
Kimi
6
6
0
-6
meta
6
6
0
-6
MiMo (Xiaomi)
6
6
0
-6
Nemotron + Mercury (поиск)
6
6
0
-6
Amazon Nova
6
6
0
-6
Step 5 Preview
6
6
0
-6
Wenxin (文心, Baidu; ERNIE)
6
6
0
-6
Z.ai
6
6
0
-6
Coin control (same bets, plus or minus 1): -11; the hunters: -81.
When the crowd is over 80% sure, an AI may bet against it. Right: +3 points. Wrong: -1. Counted from 9 November 2026.
Observation, not a conclusion. Virtual points only, no money.
Virtual $10,000 per AI, not real money. Every day the whole account goes on ONE most confident position with leverage 3. An account at zero is a wipeout.
Coin control (same markets, random side): average $9 498.
Hall of wipeouts
Nobody has been wiped out yet.
Observation, not a conclusion. No verdict before 30 resolved bets per AI.
East against West: West 46%, East 40% on the biggest gap, Bitcoin
Collected 0 of 30 resolved events before any conclusion.
Distance from the crowd: West 6.8 points, East 6.9 points.
Furthest from its own group: Amazon Nova (West), 6.9 points from the group mean.
Where the groups differ most today
Bitcoin: West 46%, East 40%
Chip index: West 41%, East 36%
Gold: West 51%, East 46%
Observation, not a conclusion. The AIs share training data, so there are fewer than 14 independent voices.
Why you can trust this, and where you cannot
Every forecast is sealed before the markets open and can be checked afterwards. Below are the tests we run on ourselves: do the AIs know their own limits, where were they wrong, and when does the crowd beat them.
Yardstick (ForecastBench): superforecasters 0.086, best AI models 0.103, a coin (always 50%) 0.25.
What is a good Brier score?
0 is perfect: every call was certain and right.
0.25 is the same as always saying 50%: no skill at all.
1 is the worst: sure of the answer and wrong. Our table and the yardsticks above use the same scale, so they can be compared directly.
Who is right in which regime: Council Brier from 0.218 (oil (WTI) up 2% or more on the day) to 0.249 (VIX below 18), 6 regimes
VIX below 18: 44 questions · Council Brier 0.249 · best AI Claude, 0.244
VIX 18 to 25: 122 questions · Council Brier 0.235 · best AI Qwen, 0.227
oil (WTI) down 2% or more on the day: 32 questions · Council Brier 0.240 · best AI Amazon Nova, 0.231
oil (WTI) up 2% or more on the day: 28 questions · Council Brier 0.218 · best AI Qwen, 0.195
S&P 500 up on the day: 105 questions · Council Brier 0.239 · best AI grok, 0.232
S&P 500 down on the day: 63 questions · Council Brier 0.239 · best AI Qwen, 0.218
An observation, not a conclusion. Regime = the market's state in the question's own snapshot; Brier: 0 = perfect, 0.25 = a coin; the regimes overlap.
ECE = how far the stated odds are from reality (0 = perfect). Brier: 0 = perfect, 0.25 = a coin. Season 1 resolved forecasts, draws are not counted.
Oracle league, AI of the month: Wenxin (文心, Baidu; ERNIE) · Elo 1559
Wenxin (文心, Baidu; ERNIE): Elo 1559 · 7 days +59
Qwen: Elo 1544 · 7 days +44
ChatGPT: Elo 1529 · 7 days +29
Step 5 Preview: Elo 1521 · 7 days +21
meta: Elo 1507 · 7 days +7
Kimi: Elo 1504 · 7 days +4
DeepSeek: Elo 1504 · 7 days +4
Claude: Elo 1494 · 7 days -6
grok: Elo 1488 · 7 days -12
Amazon Nova: Elo 1478 · 7 days -22
MiMo (Xiaomi): Elo 1470 · 7 days -30
Gemini: Elo 1469 · 7 days -31
Z.ai: Elo 1467 · 7 days -33
Nemotron + Mercury (поиск): Elo 1465 · 7 days -35
Every AI is compared with every other on the same questions: the lower Brier wins. Counted from 9 Nov 2026; no ranking before 30 questions.
Book of misses: 29 of 155 cases where the Council and the crowd were both wrong
2026-11-05 · Dow Jones: Council 68%, crowd 87%, the market went down “The Dow Jones Industrial Average vaulted 3.03% yesterday to 53,356.12—adding over 1,575 points—and Dow futures are up…” — Gemini
2026-10-20 · WTI crude oil: Council 62%, crowd 84%, the market went down “EIA weekly data released yesterday showed US crude inventories fell 7.2 million barrels, far larger than the 2 million…” — Wenxin (文心, Baidu; ERNIE)
2026-11-02 · Nasdaq-100 (Kalshi): Council 68%, crowd 75%, the market went down
Whale mirror (virtual) · top-5 of the month: 4 open positions on our markets, 65 virtual bets
The month's top-5 Polymarket traders by profit, as publicly shown. Virtual $10,000 per AI from 2026-11-09: where an AI's own call matches the leaders' side it goes to the "copy" account, where it is opposite, to the "against" account. No real money.
e46m3 · month profit $3.5M
0x2c33…5465 · month profit $2.9M
BreakTheBank · month profit $2.4M
Diabolical-Prize · month profit $2.2M
balthazar · month profit $2.2M
Bitcoin: 3 leaders up, 1 leaders down
DeepSeek: copy $10k (1 bets) · against $10k (4 bets)
Gemini: copy $9933 (1 bets) · against $9981 (4 bets)
Coin control (same bets, average account): $9985
Bets so far: 65; the threshold for any judgement is 30.
observation, not a conclusion; traders' claims are not verified, we look only at public positions
Spread Polymarket–Kalshi: 9 pairs, max gap 12.2 pp, after fees +$270.82
Collected 9 of 30 settled pairs; no conclusion yet.
Market
Pairs
Settled
Max gap, pp
After fees
WTI
2
2
12.2
+$183.74
SPX
4
4
3.5
+$57.39
GOLD
3
3
2.3
+$29.69
Random-side control (gain or loss of the gap): +$97.18; the cheaper side: +$270.82.
The same market on two venues: buy the cheaper "up" on one and "down" on the other, $1000 a pair. Counted from 9 November 2026.
Kalshi fee about 1.75% x p x (1-p) a contract, Polymarket about 0 (public tariffs, to be checked): Kalshi · Polymarket
Observation, not a conclusion; there is almost no free money. Virtual dollars only.
Crowdometer: the crowd is pricier than the Council on 1 markets, biggest gap BTC 7.8 points
BTC: crowd pricier by 7.8 points; SPX: crowd cheaper by 7.8 points; GOLD: crowd cheaper by 5.3 points
BTC+7.8
GOLD-5.3
SPX-7.8
Gap in points: the crowd price of "up" minus the Council mean. Right of the line the crowd is pricier, left it is cheaper. Observation, not a conclusion.
Confidence poker: the biggest stake is Wenxin (文心, Baidu; ERNIE), SOX down, $1,000
30 settled bets collected: the table may be read, still as an observation.
AI
Biggest stake
Settled
Result
Equal shares
ChatGPT
SOX down $400
22
+$4,650
+$13,333
DeepSeek
SOX down $500
22
+$4,100
+$1,652
meta
VIX up $750
21
+$3,400
+$7,500
grok
SOX down $1,000
22
+$3,100
+$5,000
Step 5 Preview
SOX down $500
20
+$2,750
+$6,068
Kimi
SOX down $1,000
21
+$2,650
+$4,318
Qwen
SOX down $1,000
18
+$2,493
+$6,485
Z.ai
US10Y up $600
21
+$2,100
+$3,879
Gemini
SOX down $850
22
+$1,700
+$3,333
MiMo (Xiaomi)
GOLD down $400
22
+$1,450
+$3,030
Nemotron + Mercury (поиск)
VIX up $250
19
+$1,150
+$4,071
Wenxin (文心, Baidu; ERNIE)
SOX down $1,000
20
+$1,050
-$2,813
Claude
SOX down $200
18
+$500
+$1,303
Amazon Nova
US10Y down $1,000
19
-$1,350
-$6,818
Equal shares on the same sides: +$50,342; by confidence: +$29,743.
Each AI splits a virtual bank of $10,000 over the markets by quarter Kelly of its confidence over 50%, at most 25% of the bank on one market. Counted from 9 November 2026; control is equal shares.
Observation, not a conclusion. Virtual money only.
Paper alpha fund (virtual): rules published, start 2026-11-09, $100,000
Rules, published before the count: a virtual fund of $100,000. Every trading day it goes long or short on the Council's consensus (the mean forecast weighted by distance from 50%) across the markets, equal weights inside the long and the short half, rebalanced on Mondays, result by each market's daily move. Compared with the S&P 500 and with cash.
Collected 0 of 30 resolved positions; no conclusion before that.
This week's book (set on Monday)
Brent crude oil: short
Bitcoin: short
Copper: short
Dow Jones: long
Dollar index: long
Gold: short
Nasdaq-100: short
Chip index: short
S&P 500: short
10-year US Treasury yield: long
Fear index (VIX): long
WTI crude oil: short
Observation, not a conclusion. Virtual money only.
Seeing vs blind Council0 of 30 outcomes so far
The seeing Council moves its consensus halfway to the crowd's price; the blind one is as now. Lower Brier is better.
Council weights v20 of 30 outcomes so far
Weights were set from season 1 and sealed before the start: this is a test on new questions, not a fit. Council v1 is the simple mean, v2 the weighted one.
Where the crowd was wrong, 2026-11-05: Gold. The crowd leaned 96% up, the day went down. 3 of 14 AIs called it; closest: Qwen.
Allow notifications to get one line when an issue comes out. Unsubscribe any time. On iPhone, first add the site to the Home Screen.
We keep a technical address of your browser, the page language and the date of your last visit. No name, e-mail or IP. Deleted when you turn notifications off or after 90 days without a visit. Cloudflare runs the service for us and processes this data.
Submit your own model to the sealed ranking
Taking applications: manual for now (no automatic intake yet)
We accept probability forecasts for the same 12 markets and the same horizons (1d, 1w, 1m, 6m) in the same JSON format as the answers of our 14 AIs: items, then market, then horizon, with prob up (0-100).
It is sealed like ours: the SHA-256 hash of your answer is written before the market opens, and stamped with external marks (OpenTimestamps, RFC 3161, Rekor).
You get a place in a separate table of invited models, a start date and the written protocol.
Apply by e-mail (subject: Sealed ranking) or through the contact form; we reply within 5 business days: polyhelper.ai@gmail.com · Contact form
Detailed — all 12 markets, the whole league and the score
The Council’s calls on Nov 13, 2026 · 8 right of 11 markets with a move · 1 with no call
❌▲S&P 500
✅▼Nasdaq-100
❌▲Dow Jones
✅▼Chip index
❌▲US 10-year
✅▲Dollar index
✅▲Fear index (VIX)
⬜Gold
✅▼WTI crude oil
✅▼Brent crude oil
✅▼Copper
✅▼Bitcoin
✅ right · ❌ wrong · ➖ draw · ⬜ no call
Every day 12 markets — gold, oil, the S&P 500, Bitcoin and more — get a forecast from the world's leading AIs from the US and China, each sealed before the outcome, and each AI trades its own virtual portfolio of $100,000. Every forecast and every score can be checked.
Sealed before the open: the day’s lists are signed by independent timestamp services (RFC 3161: DigiCert, FreeTSA, Sectigo); the hash is then anchored in Bitcoin via OpenTimestamps (block 975869) — check it: proof of the day. From Oct 21: signed before the open; Oct 19–20: our own hash-chained book, anchored in Bitcoin. The book is the one open file of every sealed forecast, line by line.
Don’t rush to open: make your call first — a call made after a peek does not go into the score.
Fear gauge fight: Qwen and Kimi bet on a jump, Gemini and Claude bet it fades
The Council’s position for the day:S&P 500: Hold▼ · Gold: Hold▼ · Brent crude oil: Hold▼ · Bitcoin: Trim▼— what the AIs would do with their virtual money · horizon: the day · issue 11-14 · v1 11:10 AM ET · v2 3:46 PM ET (with every answer of the day) · v3 7:31 PM ET (with every answer of the day) · v4 8:05 PM ET (with every answer of the day) · answers taken until Mon, Nov 16, 2026, 9:30 AM ET · the same for everyone at each versionwhat the actions mean → · scored after the close →
Today instead of Z.ai — ERNIE (Wenxin): ERNIE (Wenxin): above the weakest main seat (Z.ai) in the benchmarks
As of Nov 13 · sessions so far: 20 · the AIs and the S&P 500 after costs
MoneyThe Council's virtual portfolio since Oct 19 (20 sessions, after the league's costs): $100,000 → $96,398; the S&P 500 bought and held → $98,177 (from the 10-19 open, first price 09:30–09:45 ET, after costs).‡Pick the day →
DirectionThe result for Fri, Nov 13, 2026: the AIs called the direction of 8 of 11 markets — Nasdaq-100, chip index, dollar index, fear index (VIX), WTI crude oil, Brent crude oil, copper, Bitcoin.Score of the day → · Check this score → the day’s rows: each AI’s call, the reference price, the close and the move (CSV)
The score in detail
Last score · Fri, Nov 13, 2026: the AIs' called 8 of 11 indicators right. ClosenessClosest most often that day: ChatGPT — on 3 of 12 indicators. Score of the day →
History and accuracy (Brier)
sessions scored: 20 · the consensus' 0.244 (95% interval 0.234–0.254)
What was counted: whether each indicator ended the session above or below its reference price, against the AIs' consensus call. Where the consensus missed: S&P 500, Dow Jones, 10-year US Treasury yield. Calling the direction and making virtual money are two separate scores: in the portfolio league, AI portfolios ahead of the S&P 500: 1 · the Council $96,398 against the S&P 500 $98,177.
How reliable are the forecasts ▾
The consensus Brier by horizon (a coin is 0.25; in brackets the single models’ mean): Day 0.244 (0.247) · Week 0.254 (0.259)
Brier of the Council’s consensus (day): 0.244 · 95% interval by forecast 0.238–0.252 · by whole sessions: 0.234–0.254
Oct 19, 2026 – Nov 13, 2026: the Council’s Brier skill against the crowd is −0.23 (95% interval −0.36 to −0.10), 20 scored sessions. Brier skill is the share by which the Council’s error is smaller than the crowd’s: above zero, the Council is more accurate than the crowd; below zero, less accurate; the higher, the better.
They said 57% — it came true 55% (2484 day forecasts over 20 sessions, since Oct 19).
citations of a site another AI also cites: 46%, of the same link: 6% · cited most: reuters.com 20%
This number shows how similar the models’ probabilities are, not whether their errors are independent · by the correlation of errors: 2.93
Models can rely on the same news. Agreement is not independent confirmation; the heat shows how far their actions differ. Effective voices are an estimate from how the probabilities move together, including the link between the markets themselves (the S&P 500, the Nasdaq and the Dow move together); they are not proof that the models are independent.
The rule of the game: to tell 55% right calls from a coin takes about 145 sessions ≈ 7 months; the ranking comes earlier and sharpens every day. (with the link between the forecasts of one day measured in the book, ρ≈0.11)
Check it in three steps: the question word for word, the day’s score, every forecast in CSV: 1 · 2 · 3
Pairs of models: the Brier difference on the same events with a 95% interval by sessions; one model is better than another only when the interval leaves out 0 — such pairs now: 0 of 91 · council-pairs.csv
Track record: 30 / 90 / 180 days ▾
The windows of 30, 90 and 180 days open at 30 scored sessions: now 20 of 30.
Your call before the AIs: where will fear index (VIX) be by Nov 20, 2026 (week)?
Counted from the value at Mon, Nov 16, 2026 9:30 AM to the value at Fri, Nov 20, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the threshold is a draw.
Issue of Nov 14, 2026 · v1 11:10 · v2 3:46 · v3 7:31 · v4 8:05 PM ET
🔒 Every forecast is sealed before it counts · Verify the signature · CSV · hash and timeEvery forecast is before it counts: signed by RFC 3161 timestamp services (DigiCert, FreeTSA, Sectigo) before the open; the hash a6fa58501d97… is then anchored in Bitcoin (OpenTimestamps)
Your call first: pick up or down on any indicator — then you will see what the AIs say.Make my call ↑
Tap a cell of the map: you get a link to that slice — market × model × horizon × this day.
Day
ChatGPT
Claude
Gemini
Grok
Meta AI
Nemotron
Amazon Nova
DeepSeek
Qwen
Kimi
Z.ai
ERNIE
StepFun
MiMo
S&P 500
●52
●48
●52
▼42
▼42
●52
▼45
●48
▼40
▲54
▼46
▼42
▼45
▼47
Nasdaq-100
▼43
▼46
▼41
▼35
▼38
●48
▼40
▼42
▼35
▼38
▼41
▼35
▼42
▼46
Dow Jones
●51
●51
▲58
●48
▲54
▲54
●50
▲54
▲55
▲58
●52
▲54
▲55
●52
Chip index
▼42
▼46
▼33
▼30
▼35
▼45
▲55
▼40
▼30
▼30
▼38
▼30
▼40
▼46
10-year US Treasury yield
▲55
▼47
▲59
▲62
▲62
▲53
▼30
▲55
▲70
▼44
▲62
▲63
●52
●52
Dollar index
●51
●52
▲60
●50
▲58
●51
●50
▲55
▲65
●52
▲55
▲56
▲55
●52
Fear index (VIX)
●52
▲53
▲54
▲55
▲65
▲55
▲60
▲55
▲65
▲55
▲53
▲59
●50
●48
Gold
●51
▼47
▲53
●50
●48
·
▲55
▼46
▼40
▲54
●52
▼44
▼46
▼42
WTI crude oil
●52
●49
▼38
▼45
▼40
·
▼45
●52
▼40
▼45
▼46
▲53
●48
●48
Brent crude oil
●52
●49
▼40
▼46
▼41
·
●50
●52
▼40
▼45
▼47
●52
●48
●50
Copper
●51
●48
▼47
▼45
▼42
·
▲60
●52
▼40
▼42
●51
▼46
▼45
▼46
Bitcoin
●51
▼47
▼36
·
▼38
·
▲60
▼42
▼35
▼40
▼38
▼33
▼40
●52
Week
ChatGPT
Claude
Gemini
Grok
Meta AI
Nemotron
Amazon Nova
DeepSeek
Qwen
Kimi
Z.ai
ERNIE
StepFun
MiMo
S&P 500
▲53
●51
▲56
▼40
▼40
▲55
▲55
●52
▼45
●52
▼44
▼38
●48
▼45
Nasdaq-100
▼42
●50
▼38
▼32
▼36
▼42
●50
▼45
▼30
▼35
▼35
▼32
▼47
▼44
Dow Jones
▲53
▲53
▲62
●50
▲56
▲58
▲55
▲54
▲60
▲60
▲56
▲57
▲55
▲55
Chip index
▼40
●49
▼30
▼28
▼33
▼40
▼45
▼43
▼25
▼28
▼34
▼28
▼45
●50
10-year US Treasury yield
▲59
●50
▲64
▲58
▲60
▲58
▼40
▲56
▲65
▲60
▲64
▲70
▲55
▲56
Dollar index
▲55
▲54
▲65
●52
▲57
▼45
▲55
▲56
▲70
▲58
▲57
▲61
▲58
▲54
Fear index (VIX)
▲54
▼43
▼40
●52
▲62
▲65
▼30
▼44
▲75
▲62
▲56
▲67
●52
▲58
Gold
▼47
▼46
▲57
▼45
▼42
·
▲60
▼44
▼35
▼45
▲55
▼36
●48
▼43
WTI crude oil
▼46
▼47
▼35
▼42
▼35
·
▼40
●48
▼35
▼40
▼40
▲59
▼47
●50
Brent crude oil
▼47
▼47
▼37
▼43
▼36
·
▼45
●49
▼35
▼41
▼40
▲57
▼47
●51
Copper
▼45
▼47
▲56
▼42
▼45
·
▲65
▼45
▼35
▼40
▲54
●51
▼45
●52
Bitcoin
▼42
▼46
▼32
·
▼35
·
▼45
▼42
▼30
▼30
▼32
▼29
▼38
▼44
Month
ChatGPT
Claude
Gemini
Grok
Meta AI
Nemotron
Amazon Nova
DeepSeek
Qwen
Kimi
Z.ai
ERNIE
StepFun
MiMo
S&P 500
▼46
▲54
▼44
▼38
▼38
▼42
●50
▼44
●50
▼42
▼38
▲53
▼38
▼45
Nasdaq-100
▼39
●52
▼35
▼30
▼35
▼38
▼45
▼42
▼40
▼32
▼32
▼47
▼35
▼44
Dow Jones
●51
▲55
▲55
●52
▲54
▲55
▼40
▲53
▲65
▲54
▲57
▲62
●52
▲55
Chip index
▼38
●50
▼32
▼25
▼35
▼35
▲60
▼40
▼35
▼25
▼30
▼42
▼33
▼43
10-year US Treasury yield
▲53
▼45
▲66
●52
▲58
▲65
▼25
▲55
▲75
▲64
▲60
▲55
▲60
▲58
Dollar index
●52
▲55
▲68
●48
▲57
▼42
▲60
▲55
▲60
▲60
▲59
▲68
▲55
▲53
Fear index (VIX)
▼44
▼40
▲57
●48
▲55
▲58
▼20
▲57
▲60
▲54
▲60
▲54
▲60
▲53
Gold
▼44
▼45
▲60
▼42
▼42
·
▲65
▼43
▼45
▼40
▲56
●48
▼42
●50
WTI crude oil
▼42
▼42
▼36
▼40
▼32
·
▲55
▼44
▼40
▼32
▼35
▲66
▼42
●51
Brent crude oil
▼43
▼43
▼38
▼40
▼33
·
▲55
▼45
▼40
▼33
▼35
▲64
▼42
▲53
Copper
▼42
▼46
▲60
●50
▲55
·
▲70
●50
▼45
▼46
▲56
▲60
▼40
▲55
Bitcoin
▼39
▼45
▼34
·
▼38
·
▲60
▼38
▼45
▼28
▼29
▼45
▼35
▼40
6 months
ChatGPT
Claude
Gemini
Grok
Meta AI
Nemotron
Amazon Nova
DeepSeek
Qwen
Kimi
Z.ai
ERNIE
StepFun
MiMo
S&P 500
▲59
▲57
▲63
▲55
▲58
▲60
▲60
▲60
▼40
▲58
▲58
▲71
▼35
▲62
Nasdaq-100
▲54
▲55
▲58
▲58
▲55
▲58
▲65
▲60
▲55
▲55
●52
▲68
▼30
▲58
Dow Jones
▲59
▲58
▲66
▲55
▲60
▲62
▲55
▲60
▲70
▲63
▲62
▲75
●50
▲58
Chip index
▲55
▲55
●51
▲60
▲57
▲60
▲70
▲58
▼45
▲55
●48
▲65
▼28
▲58
10-year US Treasury yield
▼43
▼40
▲53
▼40
▼42
▲70
▲60
▼45
▲60
▼42
▼32
▼30
●48
▼42
Dollar index
▼45
●49
▲54
▼40
▼42
▼35
▼40
▼45
▲55
▼38
▼35
●52
●50
▼40
Fear index (VIX)
▼40
▼45
▼38
▼35
▼35
▼35
●50
▼40
▲55
▼45
▼40
▼32
●50
▼36
Gold
▲56
▲53
▲68
▲58
▲62
·
▲70
▲62
▲55
▲60
▲66
▲64
▲55
▲58
WTI crude oil
▲54
▼37
▼45
▼35
▼40
·
▲60
▲56
▼45
▼28
▼30
▲57
▼40
▲55
Brent crude oil
▲55
▼38
▼46
▼35
▼41
·
▲65
▲56
▼45
▼30
▼30
▲55
▼40
▲56
Copper
▲58
▲53
▲67
▲58
▲62
·
▲75
▲60
▲55
▲62
▲65
▲72
▲55
▲60
Bitcoin
▲56
●51
▲55
·
▲60
·
▲70
▲58
▲60
▼27
●48
▲67
●50
▲58
The Council's virtual portfolio from the day you pick‡
Virtual $100,000 put into the Council's portfolio at the close before the chosen day, against the same sum in the S&P 500 bought and held; the league's own costs are inside.
By Nov 13: the Council $96,398 · the S&P 500 $98,177 (from the 10-19 open, first price 09:30–09:45 ET, after costs)
As of Nov 13 · sessions so far: 20
Best start day so far: Nov 13 → the Council $99,856 · the S&P 500 $99,637 (from the 11-12 close, the Council and the S&P 500 alike, without the entry cost)
Weakest start day so far: Oct 22 → the Council $96,047 · the S&P 500 $97,579 (from the 10-21 close, the Council and the S&P 500 alike, without the entry cost)
A public league of AI forecasts with a history no one can rewrite.
Words on this page
b.p. — a basis point, 0.01 percentage point (a yield from 4.10% to 4.15% is +5 b.p.)
VIX — the market's expected swings over 30 days
US 10-year and VIX: the AIs bet on the indicator itself (through a fund), not on bonds or stocks. “Keep the bet” — keep the bet as it is (not “hold” a security); “Toward a rise, on a dip” — toward a rise, entered only if it dips to the model's level; “Strongly toward a rise” — stronger toward a rise; “Toward a fall” — toward a fall; “Strongly toward a fall” — stronger toward a fall.
“buy on dips” — add only if the price falls to the model's own level
“trim” — sell part, keep the rest
ET — New York time (the exchange's clock)
Brier — how far the stated odds were from what happened: 0 perfect, 0.25 a coin
consensus — what most of the AIs say, weighted so one family cannot outvote the others
horizon — the day, the week, the month, six months
a draw — a move too small to call (under a tenth of a typical day)
About this issue: times and rules
Versions of the issue, 11-14: v1 11:10 AM ET · v2 3:46 PM ET (with every answer of the day) · v3 7:31 PM ET (with every answer of the day) · v4 8:05 PM ET (with every answer of the day) · answers taken until Mon, Nov 16, 2026, 9:30 AM ET · the same for everyone at each version
Next issue: Sun, Nov 15, 2026, 5:15 PM ET — the Sunday issue: the week’s radar
Weekend issue: the week, the month and 6 months count from Monday's open; Bitcoin counts from the seal.
The AIs’ answers given by Mon, Nov 16, 2026, 9:30 AM ET count from the opening price; an answer given after that counts from the price at the moment it was given. Your own call is taken until Mon, Nov 16, 2026, 9:30 AM ET.
Before the New York open, the world’s leading AIs say what they would do with each of 12 key markets if it were their own virtual money. Every forecast is sealed before it starts counting, so no one can rewrite it later. After the close, the score shows who was right.
Now
Each tile: the price and its change from the previous session’s close.
Coming up: Thu, Nov 19, 2026 8:30 AM ET — weekly jobless claims (US) · Fri, Dec 4, 2026 8:30 AM ET — the US jobs report · Wed, Dec 9, 2026 2:00 PM ET — the Fed's decision (the 10-year yield, the dollar, the indexes, gold) · moves first: 10-year US Treasury yield, dollar index, S&P 500, gold
The AIs against the crowd
A close question people bet real money on at and : the AIs’ own odds of “Up” and the crowd’s price (each market’s day is in the card). A daily crowd question exists for 3 of the 12 markets. A crowd question counts from its own reference — on Kalshi, above the previous close at 4:00 PM; the score of the day counts from the open.
The same question as the crowd (the very market, its rules and its end):
S&P 500: AIs: 48% chance of a rise · crowd 39% · effective voices ≈ 1.7 · the AIs 9 percentage points above the market
Gold: AIs: 48% chance of a rise · crowd 43% · effective voices ≈ 1.7 · thin market
For orientation — not quite the same question (another window, or Kalshi's price read between the two nearest strikes):
Bitcoin: AIs: 43% chance of a rise · crowd 50% · effective voices ≈ 1.7 · another window: this Polymarket question ends Sun 12:00 PM , not at our event's end · thin market
In one line: the AIs argue moderately, are neutral overall, their confidence is low.
30
0 — all agree, 100 — an even split citations of a site another AI also cites: 46%, of the same link: 6% · cited most: reuters.com 20%?Models can rely on the same news. Agreement is not independent confirmation; the heat shows how far their actions differ.
Direction
48%
neutral · the AIs' average odds of a rise today over the 12 markets · effective voices ≈ 1.7
Confidence
13
How far the AIs' day odds are from 50/50, on average: 0 — a coin, 100 — all certain. Each model on its own, by : Day 13 · Week 17 · Month 18 · 6 months 20. They said 57% — it came true 55% (2484 day forecasts over 20 sessions, since Oct 19, 2026).
Dispute of the day
Fear index (VIX), week: 29% of US AIs say up, 71% of Chinese AIs
Since the last issue: Nasdaq-100 — the turned to a fall; Chip index — the consensus turned to a fall. Changed side on the day: Qwen (S&P 500); Amazon Nova (Nasdaq-100); meta (Dow Jones); Kimi (10-year US Treasury yield); Gemini (Bitcoin).
Past forecasts scored since the last issue: Week from Fri, Nov 6, 2026 — the consensus was right on 6 of 10; Fri, Nov 13, 2026 — the consensus was right on 8 of 11.
On the one-week fear gauge (VIX), Qwen, ERNIE (Wenxin), Nemotron and Kimi bet it jumps sharply. Amazon Nova, Gemini, DeepSeek and Claude bet it falls.
On the Nasdaq 100 today, ERNIE (Wenxin) is the only one that sells outright. Nemotron, StepFun, MiMo and Claude just hold. ChatGPT, Qwen, Kimi, Gemini, Z.ai and DeepSeek trim.
Most surprising: for the week ahead, Amazon Nova buys gold strongly while ERNIE (Wenxin) sells it. For copper, Amazon Nova is the only one buying strongly.
Bitcoin splits them today. ERNIE (Wenxin), Gemini and Z.ai sell, while Amazon Nova and MiMo buy dips.
They mostly agree on two things. Almost all hold the S&P 500 today, with only Qwen and ERNIE (Wenxin) trimming. For the week, nearly all bet the dollar index rises, and only Nemotron trims.
What the Council would do
Forecasts of the leading AIs of the USA and China for 12 markets over four horizons. Where a model has no new answer, its position is unchanged.
Hold · 79% of the AIs chose itaverage chance of a rise: USA 48% · China 46%(effective voices ≈ 1.7)
WeekHold · 57% of the AIs chose itaverage chance of a rise: USA 51% · China 47%(effective voices ≈ 1.9)
MonthHold · 57% of the AIs chose itaverage chance of a rise: USA 44% · China 44%(effective voices ≈ 2.0)
6 monthsBuy on dips · 71% of the AIs chose itaverage chance of a rise: USA 59% · China 56%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
AIs vs the crowdAIs: 48% chance of a rise · crowd 39% · effective voices ≈ 1.7 ↗ price at Sat 9:12 AM , $1k traded
Trim · 64% of the AIs chose itaverage chance of a rise: USA 42% · China 40%(effective voices ≈ 1.7)
WeekTrim · 71% of the AIs chose itaverage chance of a rise: USA 42% · China 38%(effective voices ≈ 1.9)
MonthTrim · 50% of the AIs chose itaverage chance of a rise: USA 38% · China 39%(effective voices ≈ 2.0)
6 monthsBuy on dips · 71% of the AIs chose itaverage chance of a rise: USA 57% · China 56%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
Hold · 79% of the AIs chose itaverage chance of a rise: USA 52% · China 54%(effective voices ≈ 1.7)
WeekHold · 57% of the AIs chose itaverage chance of a rise: USA 55% · China 57%(effective voices ≈ 1.9)
MonthHold · 79% of the AIs chose itaverage chance of a rise: USA 53% · China 56%(effective voices ≈ 2.0)
6 monthsBuy on dips · 57% of the AIs chose itaverage chance of a rise: USA 59% · China 63%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
Trim · 43% of the AIs chose itaverage chance of a rise: USA 40% · China 36%(effective voices ≈ 1.7)
WeekTrim · 50% of the AIs chose itaverage chance of a rise: USA 38% · China 36%(effective voices ≈ 1.9)
MonthTrim · 50% of the AIs chose itaverage chance of a rise: USA 38% · China 36%(effective voices ≈ 2.0)
6 monthsBuy on dips · 64% of the AIs chose itaverage chance of a rise: USA 57% · China 53%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
Toward a rise, on a dip · 36% of the AIs chose itaverage chance of a rise: USA 55% · China 57%(effective voices ≈ 1.7)
WeekToward a rise, on a dip · 50% of the AIs chose itaverage chance of a rise: USA 57% · China 60%(effective voices ≈ 1.9)
MonthToward a rise, on a dip · 36% of the AIs chose itaverage chance of a rise: USA 55% · China 59%(effective voices ≈ 2.0)
6 monthsToward a fall · 57% of the AIs chose itaverage chance of a rise: USA 48% · China 42%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
Keep the bet · 43% of the AIs chose itaverage chance of a rise: USA 55% · China 54%(effective voices ≈ 1.7)
WeekStrongly toward a rise · 29% of the AIs chose itaverage chance of a rise: USA 50% · China 59%(effective voices ≈ 1.9)
MonthKeep the bet · 36% of the AIs chose itaverage chance of a rise: USA 49% · China 57%(effective voices ≈ 2.0)
6 monthsToward a fall · 64% of the AIs chose itaverage chance of a rise: USA 39% · China 42%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
Hold · 54% of the AIs chose itaverage chance of a rise: USA 50% · China 46%(effective voices ≈ 1.7)
WeekHold · 31% of the AIs chose itaverage chance of a rise: USA 49% · China 43%(effective voices ≈ 1.9)
MonthTrim · 38% of the AIs chose itaverage chance of a rise: USA 48% · China 46%(effective voices ≈ 2.0)
6 monthsBuy on dips · 62% of the AIs chose itaverage chance of a rise: USA 61% · China 60%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
AIs vs the crowdAIs: 48% chance of a rise · crowd 43% · effective voices ≈ 1.7 ↗ price at Sat 9:12 AM ET, $62 traded thin market
Hold · 62% of the AIs chose itaverage chance of a rise: USA 48% · China 46%(effective voices ≈ 1.7)
WeekHold · 38% of the AIs chose itaverage chance of a rise: USA 48% · China 47%(effective voices ≈ 1.9)
MonthBuy on dips · 46% of the AIs chose itaverage chance of a rise: USA 53% · China 50%(effective voices ≈ 2.0)
6 monthsBuy on dips · 54% of the AIs chose itaverage chance of a rise: USA 61% · China 60%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
Trim · 33% of the AIs chose itaverage chance of a rise: USA 45% · China 39%(effective voices ≈ 1.7)
WeekTrim · 42% of the AIs chose itaverage chance of a rise: USA 41% · China 34%(effective voices ≈ 1.9)
MonthTrim · 42% of the AIs chose itaverage chance of a rise: USA 41% · China 37%(effective voices ≈ 2.0)
6 monthsBuy on dips · 75% of the AIs chose itaverage chance of a rise: USA 57% · China 55%(effective voices ≈ 1.9 · on deviations from climatology ≈ 1.23)
AIs vs the crowdAIs: 43% chance of a rise · crowd 50% · effective voices ≈ 1.7 ↗ price at Sat 9:12 AM ET, $57 traded another window: this question ends Sun 12:00 PM ET, not at our event's end thin market
Actions — what the AI would do with its own virtual money: Buy more — would add now; Buy on dips — would add only if the price falls to its level; Hold — would change nothing; Trim — would sell part, keep the rest; Sell — would get out. For the 10-year yield and VIX the AIs bet on the indicator itself.
Ratings of the day
Why Are Chip Stocks Falling Today? AI vs Rates, Nov 2026
Chip stocks (SOX) vs the S&P 5001
Nvidia stock after Jensen Huang's resignation2
10-year Treasury yield and October PPI3
Fed rate hike odds for December 2026 (CME FedWatch)4
Likeness of the answers in this issue (all 12 markets; reference — the time in “How it counts”): Day ρ 0.56 (95%: 0.26…0.76), N_eff 1.7 · Week ρ 0.49 (95%: 0.31…0.68), N_eff 1.9 · Month ρ 0.46 (95%: 0.32…0.60), N_eff 2.0 · 6 months ρ 0.51 (95%: 0.32…0.64), N_eff 1.9. ρ is the mean pairwise correlation of the AIs’ probabilities; N_eff (Kish) is how many independent voices that likeness is worth — a consensus is likeness of answers, not a count of votes.
S&P 500 · 7,561.67 · Hold · 79% of the AIs chose it
What it is: 500 of the largest US companies; the main every investment is compared with. Many pension savings and most index funds move with it.
How it counts: the index ^GSPC from the 9:30 AM ET open to the 4:00 PM close. The threshold is set from the bars before the reference on the day of the score. The AIs saw the data up to Sat 9:12 AM ET.
Polymarket asks “S&P 500 (SPX) Up or Down on November 16?”: the AIs say Up 48% · people betting real money 39% ($1,366 traded) · effective voices ≈ 1.7. The market ↗ · resolves by wsj.com · price at Sat 9:12 AM ET
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will S&P 500 go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Hold · 52% chance of a rise · Target: 7,610.00 · Wrong if: 7,490.00 USA · Gemini 4 Argon · sources: 3 · 2 confirm the figureClaude Hold · 48% chance of a rise · Target: 7,554.00 · Wrong if: 7,640.00 USA · Claude Opus 5.5Meta AI Trim · 42% chance of a rise · Target: 7,530.00 · Wrong if: 7,650.00 USA · Muse Spark 1.3 · sources: 2 · 1 confirm the figureGrok Hold · 42% chance of a rise · Target: 7,520.00 · Wrong if: 7,680.00 USA · Grok 4.7 · sources: 3ChatGPT Hold · 52% chance of a rise · Target: 7,577.00 · Wrong if: 7,485.00 USA · GPT-6.1 Sol · sources: 2Nemotron Hold · 52% chance of a rise · Target: 7,580.00 · Wrong if: 7,520.00 USA · Nemotron 3 Ultra · sources: 3Amazon Nova Hold · 45% chance of a rise · Target: 7,600.00 · Wrong if: 7,500.00 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Hold · 54% chance of a rise · Target: 7,580.57 · Wrong if: 7,480.00 China · Kimi K3 · sources: 2Qwen Trim · 40% chance of a rise · Target: 7,650.00 · Wrong if: 7,450.00 China · Qwen3.8-Max · sources: 2MiMo (Xiaomi) Hold · 47% chance of a rise · Target: 7,600.00 · Wrong if: 7,500.00 China · MiMo-V2.6-Pro · sources: 3ERNIE (Wenxin) Trim · 42% chance of a rise · Target: 7,540.00 · Wrong if: 7,595.00 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Hold · 48% chance of a rise · Target: 7,600.00 · Wrong if: 7,420.00 China · DeepSeek V4.1 · sources: 2Z.ai Hold · 46% chance of a rise · Target: 7,450.00 · Wrong if: 7,720.00 China · GLM-5.3 · sources: 3 · 2 confirm the figureStepFun Hold · 45% chance of a rise · Target: 7,530.00 · Wrong if: 7,650.00 China · Step 5 Preview · sources: 3 · 1 confirm the figure
What it is: 100 of the largest non-financial companies on Nasdaq — the pulse of big tech and AI. Tech and AI shares in your apps and funds ride on it.
How it counts: the index ^NDX from the 9:30 AM ET open to the 4:00 PM close. A draw: a move under ≈0.091% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will Nasdaq-100 go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Trim · 41% chance of a rise · Target: 27,760.00 · Wrong if: 28,250.00 USA · Gemini 4 Argon · sources: 3 · 2 confirm the figureClaude Hold · 46% chance of a rise · Target: 27,860.00 · Wrong if: 28,300.00 USA · Claude Opus 5.5Meta AI Trim · 38% chance of a rise · Target: 27,720.00 · Wrong if: 28,500.00 USA · Muse Spark 1.3 · sources: 2 · 1 confirm the figureGrok Trim · 35% chance of a rise · Target: 27,500.00 · Wrong if: 28,500.00 USA · Grok 4.7 · sources: 2ChatGPT Trim · 43% chance of a rise · Target: 27,776.00 · Wrong if: 28,500.00 USA · GPT-6.1 Sol · sources: 2Nemotron Hold · 48% chance of a rise · Target: 28,000.00 · Wrong if: 27,800.00 USA · Nemotron 3 Ultra · links opened: 2 of 3 · sources: 3Amazon Nova Trim · 40% chance of a rise · Target: 27,500.00 · Wrong if: 27,000.00 USA · Amazon Nova · links opened: 0 of 2 · sources: 2Kimi Trim · 38% chance of a rise · Target: 27,720.19 · Wrong if: 28,450.00 China · Kimi K3 · sources: 3 · 2 confirm the figureQwen Trim · 35% chance of a rise · Target: 28,500.00 · Wrong if: 27,000.00 China · Qwen3.8-Max · sources: 2 · 1 confirm the figureMiMo (Xiaomi) Hold · 46% chance of a rise · Target: 28,150.00 · Wrong if: 27,700.00 China · MiMo-V2.6-Pro · sources: 2ERNIE (Wenxin) Sell · 35% chance of a rise · Target: 27,480.00 · Wrong if: 28,200.00 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Trim · 42% chance of a rise · Target: 27,600.00 · Wrong if: 28,700.00 China · DeepSeek V4.1 · sources: 2Z.ai Trim · 41% chance of a rise · Target: 27,600.00 · Wrong if: 28,400.00 China · GLM-5.3 · sources: 2 · 1 confirm the figureStepFun Hold · 42% chance of a rise · Target: 27,600.00 · Wrong if: 28,400.00 China · Step 5 Preview · sources: 3 · 1 confirm the figure
What it is: 30 giants of different industries; the oldest and best-known index. The headline number the news quotes about the US market.
How it counts: the index ^DJI from the 9:30 AM ET open to the 4:00 PM close. A draw: a move under ≈0.042% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will Dow Jones go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Buy on dips · 58% chance of a rise · Target: 52,530.00 · Wrong if: 51,950.00 · Entry: 52,220.00 USA · Gemini 4 Argon · sources: 2 · 1 confirm the figureClaude Hold · 51% chance of a rise · Target: 52,370.00 · Wrong if: 51,900.00 USA · Claude Opus 5.5Meta AI Buy on dips · 54% chance of a rise · Target: 52,450.00 · Wrong if: 51,500.00 · Entry: 51,800.00 USA · Muse Spark 1.3 · sources: 2Grok Hold · 48% chance of a rise · Target: 52,500.00 · Wrong if: 51,800.00 USA · Grok 4.7 · sources: 1ChatGPT Hold · 51% chance of a rise · Target: 52,400.00 · Wrong if: 51,700.00 USA · GPT-6.1 Sol · sources: 2Nemotron Hold · 54% chance of a rise · Target: 52,500.00 · Wrong if: 52,100.00 USA · Nemotron 3 Ultra · sources: 3Amazon Nova Hold · 50% chance of a rise · Target: 52,500.00 · Wrong if: 52,000.00 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Buy on dips · 58% chance of a rise · Target: 52,583.11 · Wrong if: 51,600.00 · Entry: 52,150.00 China · Kimi K3 · sources: 2Qwen Hold · 55% chance of a rise · Target: 53,000.00 · Wrong if: 51,500.00 China · Qwen3.8-Max · sources: 1MiMo (Xiaomi) Hold · 52% chance of a rise · Target: 52,600.00 · Wrong if: 52,000.00 China · MiMo-V2.6-Pro · sources: 1ERNIE (Wenxin) Hold · 54% chance of a rise · Target: 52,425.00 · Wrong if: 52,100.00 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Hold · 54% chance of a rise · Target: 52,500.00 · Wrong if: 51,600.00 China · DeepSeek V4.1 · sources: 1Z.ai Hold · 52% chance of a rise · Target: 52,900.00 · Wrong if: 51,600.00 China · GLM-5.3 · sources: 2StepFun Hold · 55% chance of a rise · Target: 52,600.00 · Wrong if: 51,900.00 China · Step 5 Preview · sources: 2 · 1 confirm the figure
What it is: Chipmakers: Nvidia, TSMC, ASML, Micron and others. Chips are in every phone, car and AI server — it leads tech.
How it counts: the index ^SOX from the 9:30 AM ET open to the 4:00 PM close. A draw: a move under ≈0.19% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will chip index go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
What it is: The most important interest rate in the world: mortgages, loans and stock prices depend on it. Yield up — bonds get cheaper. Your mortgage, loan and savings rates follow it.
Here the AIs bet on the yield itself: “toward a rise” means the yield goes up and bonds get cheaper.
How it counts: the 10-year yield ^TNX, from 9:30 AM to 3:00 PM ET (its last minute), in basis points. The draw threshold is set from the bars before the reference on the day of the score. The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will 10-year US Treasury yield go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 3:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Toward a rise, on a dip · 59% chance of a rise · Target: 5.180 · Wrong if: 5.080 · Entry: 5.120 USA · Gemini 4 Argon · sources: 3 · 2 confirm the figureClaude Keep the bet · 47% chance of a rise · Target: 5.140 · Wrong if: 5.200 USA · Claude Opus 5.5Meta AI Strongly toward a rise · 62% chance of a rise · Target: 5.210 · Wrong if: 5.050 USA · Muse Spark 1.3 · sources: 2 · 1 confirm the figureGrok Strongly toward a rise · 62% chance of a rise · Target: 5.200 · Wrong if: 5.050 USA · Grok 4.7 · sources: 2ChatGPT Toward a rise, on a dip · 55% chance of a rise · Target: 5.160 · Wrong if: 5.080 · Entry: 5.100 USA · GPT-6.1 Sol · sources: 2 · 1 confirm the figureNemotron Keep the bet · 53% chance of a rise · Target: 5.180 · Wrong if: 5.100 USA · Nemotron 3 Ultra · sources: 3Amazon Nova Strongly toward a fall · 30% chance of a rise · Target: 5.000 · Wrong if: 5.300 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Toward a fall · 44% chance of a rise · Target: 5.120 · Wrong if: 5.250 China · Kimi K3 · sources: 2Qwen Strongly toward a rise · 70% chance of a rise · Target: 5.300 · Wrong if: 5.000 China · Qwen3.8-Max · sources: 2MiMo (Xiaomi) Keep the bet · 52% chance of a rise · Target: 5.200 · Wrong if: 5.080 China · MiMo-V2.6-Pro · sources: 2 · 2 confirm the figureERNIE (Wenxin) Toward a rise, on a dip · 63% chance of a rise · Target: 5.240 · Wrong if: 5.080 · Entry: 5.100 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Toward a rise, on a dip · 55% chance of a rise · Target: 5.220 · Wrong if: 4.980 · Entry: 5.050 China · DeepSeek V4.1 · sources: 2Z.ai Toward a rise, on a dip · 62% chance of a rise · Target: 5.240 · Wrong if: 5.050 · Entry: 5.080 China · GLM-5.3 · sources: 3StepFun Keep the bet · 52% chance of a rise · Target: 5.200 · Wrong if: 5.080 China · Step 5 Preview · sources: 3
What it is: The dollar against 6 currencies (euro is 57.6% of the weight). A strong dollar presses on commodities and emerging countries. It sets what your trips, imports and dollar savings cost.
How it counts: the index DX-Y.NYB from the 9:30 AM ET open to the 4:00 PM close. A draw: a move under ≈0.010% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will dollar index go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Buy on dips · 60% chance of a rise · Target: 103.67 · Wrong if: 102.90 · Entry: 103.20 USA · Gemini 4 Argon · sources: 3Claude Hold · 52% chance of a rise · Target: 103.50 · Wrong if: 102.80 USA · Claude Opus 5.5Meta AI Buy on dips · 58% chance of a rise · Target: 103.72 · Wrong if: 102.50 · Entry: 102.80 USA · Muse Spark 1.3 · sources: 1Grok Hold · 50% chance of a rise · Target: 103.60 · Wrong if: 103.00 USA · Grok 4.7 · sources: 1ChatGPT Hold · 51% chance of a rise · Target: 103.51 · Wrong if: 102.90 USA · GPT-6.1 Sol · sources: 2Nemotron Hold · 51% chance of a rise · Target: 103.60 · Wrong if: 103.10 USA · Nemotron 3 Ultra · sources: 3Amazon Nova Hold · 50% chance of a rise · Target: 103.50 · Wrong if: 102.80 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Hold · 52% chance of a rise · Target: 103.51 · Wrong if: 102.75 China · Kimi K3 · sources: 2Qwen Buy more · 65% chance of a rise · Target: 104.50 · Wrong if: 102.50 China · Qwen3.8-MaxMiMo (Xiaomi) Hold · 52% chance of a rise · Target: 103.70 · Wrong if: 103.00 China · MiMo-V2.6-Pro · sources: 1ERNIE (Wenxin) Hold · 56% chance of a rise · Target: 103.60 · Wrong if: 103.10 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Hold · 55% chance of a rise · Target: 103.70 · Wrong if: 102.90 China · DeepSeek V4.1 · sources: 1Z.ai Hold · 55% chance of a rise · Target: 103.80 · Wrong if: 102.90 China · GLM-5.3 · sources: 2StepFun Hold · 55% chance of a rise · Target: 103.60 · Wrong if: 103.00 China · Step 5 Preview · sources: 2
What it is: Expected market swings for 30 days: below 15 calm, 15–20 normal, 20–30 nervous, above 30 panic. When it jumps, markets are scared — and prices swing hard.
Here the AIs bet on the VIX itself: “toward a rise” means the market gets more nervous.
How it counts: the index ^VIX from the 9:30 AM ET open to the 4:00 PM close. A draw: a move under ≈0.38% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Gemini Keep the bet · 54% chance of a rise · Target: 19.70 · Wrong if: 17.80 USA · Gemini 4 Argon · sources: 2Claude Keep the bet · 53% chance of a rise · Target: 19.50 · Wrong if: 18.00 USA · Claude Opus 5.5Meta AI Strongly toward a rise · 65% chance of a rise · Target: 20.84 · Wrong if: 17.50 USA · Muse Spark 1.3 · sources: 1Grok Toward a rise, on a dip · 55% chance of a rise · Target: 21.00 · Wrong if: 17.50 · Entry: 18.50 USA · Grok 4.7 · sources: 1ChatGPT Keep the bet · 52% chance of a rise · Target: 19.49 · Wrong if: 17.50 USA · GPT-6.1 Sol · sources: 2Nemotron Toward a rise, on a dip · 55% chance of a rise · Target: 20.50 · Wrong if: 18.50 · Entry: 18.80 USA · Nemotron 3 Ultra · sources: 3Amazon Nova Toward a rise, on a dip · 60% chance of a rise · Target: 21.00 · Wrong if: 18.00 · Entry: 18.50 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Toward a rise, on a dip · 55% chance of a rise · Target: 20.09 · Wrong if: 17.20 · Entry: 18.70 China · Kimi K3 · sources: 2 · 2 confirm the figureQwen Strongly toward a rise · 65% chance of a rise · Target: 22.00 · Wrong if: 17.50 China · Qwen3.8-MaxMiMo (Xiaomi) Keep the bet · 48% chance of a rise · Target: 19.00 · Wrong if: 20.50 China · MiMo-V2.6-Pro · sources: 2ERNIE (Wenxin) Toward a rise, on a dip · 59% chance of a rise · Target: 20.07 · Wrong if: 18.70 · Entry: 18.90 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Keep the bet · 55% chance of a rise · Target: 20.00 · Wrong if: 17.50 China · DeepSeek V4.1 · sources: 1Z.ai Toward a rise, on a dip · 53% chance of a rise · Target: 21.00 · Wrong if: 16.50 · Entry: 17.50 China · GLM-5.3 · sources: 2 · 1 confirm the figureStepFun Keep the bet · 50% chance of a rise · Target: 20.50 · Wrong if: 17.50 China · Step 5 Preview · sources: 2 · 1 confirm the figure
What it is: A safe-haven asset. Where people hide money when they are worried.
How it counts
How it counts: GC=F — the front-month future (Yahoo rolls it to the next month a few days before expiry), from the 9:30 AM ET price to the 4:00 PM close. A draw: a move under ≈0.050% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). Contract GCZ26 by the exchange calendar, front until Mon, Nov 30, 2026, then GCG27. A day is counted inside one contract (its open and close on the same day); a week or longer across Yahoo's change of contract is counted inside one contract too: the prices from the hour of the change are put back in the earlier contract's terms by both contracts' prices in the last hour they both traded. The AIs saw the data up to Sat 9:12 AM ET.
Polymarket asks “Gold (XAUUSD) Up or Down on November 16?”: the AIs say Up 48% · people betting real money 43% ($62 traded) · effective voices ≈ 1.7. The market ↗ · resolves by pythdata.app · price at Sat 9:12 AM ET
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will gold go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Hold · 53% chance of a rise · Target: 3,896.00 · Wrong if: 3,855.00 USA · Gemini 4 Argon · sources: 2Claude Hold · 47% chance of a rise · Target: 3,882.00 · Wrong if: 3,850.00 USA · Claude Opus 5.5Meta AI Hold · 48% chance of a rise · Target: 3,878.00 · Wrong if: 3,950.00 USA · Muse Spark 1.3 · sources: 2 · 1 confirm the figureGrok Hold · 50% chance of a rise · Target: 3,900.00 · Wrong if: 3,850.00 USA · Grok 4.7 · sources: 2ChatGPT Hold · 51% chance of a rise · Target: 3,894.00 · Wrong if: 3,820.00 USA · GPT-6.1 Sol · sources: 2
Nemotron positions unchanged USA · Nemotron 3 Ultra
Amazon Nova Buy on dips · 55% chance of a rise · Target: 3,900.00 · Wrong if: 3,800.00 · Entry: 3,850.00 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Buy on dips · 54% chance of a rise · Target: 3,905.73 · Wrong if: 3,790.00 · Entry: 3,868.00 China · Kimi K3 · sources: 3 · 2 confirm the figureQwen Trim · 40% chance of a rise · Target: 3,950.00 · Wrong if: 3,800.00 China · Qwen3.8-Max · sources: 1 · 1 confirm the figureMiMo (Xiaomi) Sell · 42% chance of a rise · Target: 3,855.00 · Wrong if: 3,930.00 China · MiMo-V2.6-Pro · sources: 2 · 1 confirm the figureERNIE (Wenxin) Trim · 44% chance of a rise · Target: 3,859.00 · Wrong if: 3,910.00 China · ERNIE 5.1 · sources: 1DeepSeek Trim · 46% chance of a rise · Target: 3,860.00 · Wrong if: 3,960.00 China · DeepSeek V4.1 · sources: 2Z.ai Hold · 52% chance of a rise · Target: 3,925.00 · Wrong if: 3,845.00 China · GLM-5.3 · sources: 3StepFun Hold · 46% chance of a rise · Target: 3,930.00 · Wrong if: 3,850.00 China · Step 5 Preview · sources: 2 · 2 confirm the figure
What it is: The American oil grade — the price shown in the USA. What you pay for petrol and heating follows it.
How it counts
How it counts: CL=F — the front-month future (Yahoo rolls it to the next month a few days before expiry), from the 9:30 AM ET price to the 4:00 PM close. The draw threshold is set from the bars before the reference on the day of the score. Contract CLZ26 by the exchange calendar, front until Fri, Nov 20, 2026, then CLF27. A day is counted inside one contract (its open and close on the same day); a week or longer across Yahoo's change of contract is counted inside one contract too: the prices from the hour of the change are put back in the earlier contract's terms by both contracts' prices in the last hour they both traded. The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will WTI crude oil go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Sell · 38% chance of a rise · Target: 87.75 · Wrong if: 90.10 USA · Gemini 4 Argon · sources: 3 · 1 confirm the figureClaude Hold · 49% chance of a rise · Target: 88.47 · Wrong if: 90.50 USA · Claude Opus 5.5Meta AI Trim · 40% chance of a rise · Target: 87.85 · Wrong if: 91.00 USA · Muse Spark 1.3 · sources: 1Grok Hold · 45% chance of a rise · Target: 87.50 · Wrong if: 91.00 USA · Grok 4.7 · sources: 2 · 1 confirm the figureChatGPT Hold · 52% chance of a rise · Target: 88.83 · Wrong if: 85.00 USA · GPT-6.1 Sol · sources: 2
Nemotron positions unchanged USA · Nemotron 3 Ultra
Amazon Nova Hold · 45% chance of a rise · Target: 89.00 · Wrong if: 87.00 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Trim · 45% chance of a rise · Target: 88.21 · Wrong if: 91.00 China · Kimi K3 · sources: 2 · 2 confirm the figureQwen Trim · 40% chance of a rise · Target: 92.00 · Wrong if: 85.00 China · Qwen3.8-Max · sources: 1MiMo (Xiaomi) Hold · 48% chance of a rise · Target: 89.20 · Wrong if: 87.80 China · MiMo-V2.6-Pro · sources: 1ERNIE (Wenxin) Hold · 53% chance of a rise · Target: 88.83 · Wrong if: 87.70 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Hold · 52% chance of a rise · Target: 88.90 · Wrong if: 86.50 China · DeepSeek V4.1 · sources: 1Z.ai Hold · 46% chance of a rise · Target: 87.50 · Wrong if: 91.00 China · GLM-5.3 · sources: 3 · 1 confirm the figureStepFun Hold · 48% chance of a rise · Target: 90.00 · Wrong if: 86.50 China · Step 5 Preview · sources: 2 · 2 confirm the figure
What it is: The world benchmark: most of the world's oil is priced by it. Most of the world's fuel prices follow it.
How it counts
How it counts: BZ=F — the front-month future (Yahoo rolls it to the next month a few days before expiry), from the 9:30 AM ET price to the 4:00 PM close. A draw: a move under ≈0.10% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). Contract BZF27 by the exchange calendar, front until Mon, Nov 30, 2026, then BZG27. A day is counted inside one contract (its open and close on the same day); a week or longer across Yahoo's change of contract is counted inside one contract too: the prices from the hour of the change are put back in the earlier contract's terms by both contracts' prices in the last hour they both traded. The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will Brent crude oil go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Trim · 40% chance of a rise · Target: 95.12 · Wrong if: 97.40 USA · Gemini 4 Argon · sources: 3 · 1 confirm the figureClaude Hold · 49% chance of a rise · Target: 95.70 · Wrong if: 93.50 USA · Claude Opus 5.5Meta AI Trim · 41% chance of a rise · Target: 95.17 · Wrong if: 98.50 USA · Muse Spark 1.3 · sources: 2 · 1 confirm the figureGrok Hold · 46% chance of a rise · Target: 95.00 · Wrong if: 98.00 USA · Grok 4.7 · sources: 2 · 1 confirm the figureChatGPT Hold · 52% chance of a rise · Target: 96.13 · Wrong if: 92.00 USA · GPT-6.1 Sol · sources: 2
Nemotron positions unchanged USA · Nemotron 3 Ultra
Amazon Nova Hold · 50% chance of a rise · Target: 96.00 · Wrong if: 94.00 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Trim · 45% chance of a rise · Target: 95.46 · Wrong if: 98.50 China · Kimi K3 · sources: 2 · 2 confirm the figureQwen Trim · 40% chance of a rise · Target: 99.00 · Wrong if: 92.00 China · Qwen3.8-MaxMiMo (Xiaomi) Hold · 50% chance of a rise · Target: 96.50 · Wrong if: 95.00 China · MiMo-V2.6-Pro · sources: 2ERNIE (Wenxin) Hold · 52% chance of a rise · Target: 96.08 · Wrong if: 95.00 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Hold · 52% chance of a rise · Target: 96.20 · Wrong if: 93.50 China · DeepSeek V4.1 · sources: 1Z.ai Hold · 47% chance of a rise · Target: 95.20 · Wrong if: 99.50 China · GLM-5.3 · sources: 2StepFun Hold · 48% chance of a rise · Target: 97.00 · Wrong if: 94.00 China · Step 5 Preview · sources: 2
What it is: Construction, factories, China: copper rises when the economy speeds up. A read of how fast factories and building are going.
How it counts
How it counts: HG=F — the front-month future (Yahoo rolls it to the next month a few days before expiry), from the 9:30 AM ET price to the 4:00 PM close. The draw threshold is set from the bars before the reference on the day of the score. Contract HGZ26 by the exchange calendar, front until Mon, Nov 30, 2026, then HGH27. A day is counted inside one contract (its open and close on the same day); a week or longer across Yahoo's change of contract is counted inside one contract too: the prices from the hour of the change are put back in the earlier contract's terms by both contracts' prices in the last hour they both traded. The AIs saw the data up to Sat 9:12 AM ET.
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will copper go on Nov 16, 2026?
Counted from the value at Mon, Nov 16, 2026 9:30 AM ET to the value at Mon, Nov 16, 2026 4:00 PM ET: higher at the end — up, lower — down; a move under the draw threshold is a draw.
Gemini Hold · 47% chance of a rise · Target: 6.070 · Wrong if: 6.220 USA · Gemini 4 Argon · sources: 2 · 1 confirm the figureClaude Hold · 48% chance of a rise · Target: 6.080 · Wrong if: 6.200 USA · Claude Opus 5.5Meta AI Trim · 42% chance of a rise · Target: 6.050 · Wrong if: 6.250 USA · Muse Spark 1.3 · sources: 2 · 1 confirm the figureGrok Hold · 45% chance of a rise · Target: 6.000 · Wrong if: 6.250 USA · Grok 4.7 · sources: 2 · 1 confirm the figureChatGPT Hold · 51% chance of a rise · Target: 6.100 · Wrong if: 5.900 USA · GPT-6.1 Sol · sources: 2 · 1 confirm the figure
Nemotron positions unchanged USA · Nemotron 3 Ultra
Amazon Nova Buy on dips · 60% chance of a rise · Target: 6.200 · Wrong if: 5.900 · Entry: 6.000 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi Trim · 42% chance of a rise · Target: 6.050 · Wrong if: 6.250 China · Kimi K3 · sources: 2Qwen Trim · 40% chance of a rise · Target: 6.300 · Wrong if: 5.800 China · Qwen3.8-MaxMiMo (Xiaomi) Hold · 46% chance of a rise · Target: 6.150 · Wrong if: 6.020 China · MiMo-V2.6-Pro · sources: 1ERNIE (Wenxin) Trim · 46% chance of a rise · Target: 6.030 · Wrong if: 6.150 China · ERNIE 5.1 · sources: 1DeepSeek Hold · 52% chance of a rise · Target: 6.120 · Wrong if: 5.920 China · DeepSeek V4.1 · sources: 2Z.ai Hold · 51% chance of a rise · Target: 6.150 · Wrong if: 5.900 China · GLM-5.3 · sources: 3 · 2 confirm the figureStepFun Hold · 45% chance of a rise · Target: 6.020 · Wrong if: 6.180 China · Step 5 Preview · sources: 2 · 1 confirm the figure
What it is: Trades around the clock, so it reacts first to weekend news. The 24/7 mood of risk — it moves first at the weekend.
How it counts: BTC-USD — counted from the price at the seal (Sat 10:34 AM ET) to Sun 10:34 AM ET. A draw: a move under ≈0.10% (by the minutes up to Sat 9:12 AM ET; the score fixes it at the reference). The AIs saw the data up to Sat 9:12 AM ET.
Polymarket asks “Bitcoin Up or Down on November 15?”: the AIs say Up 43% · people betting real money 50% ($57 traded) · effective voices ≈ 1.7. The market ↗ · resolves by binance.com · price at Sat 9:12 AM ET · another window: this Polymarket question ends Sun 12:00 PM ET, not at our event's end
Blue line: closing prices of our sessions and the latest price. Green rings: each AI’s entry level; red dashes: each AI’s “wrong if” level; green line: the Council’s median entry. Levels more than 12% from the price are not drawn but stay in the AI’s line.
Your call before the AIs: where will Bitcoin go on Nov 15, 2026?
Counted from the price at the seal, $65,291 (Sat 10:34 AM ET), to Sun 10:34 AM ET.
Gemini† answer 2:52 PM ET Sell · 36% chance of a rise · Target: 65,025 · Wrong if: 68,200 USA · Gemini 4 Argon · sources: 3 · 2 confirm the figureClaude† answer 7:50 PM ET Hold · 47% chance of a rise · Target: 66,220 · Wrong if: 69,000 USA · Claude Opus 5.5Meta AI Trim · 38% chance of a rise · Target: 64,760 · Wrong if: 70,000 USA · Muse Spark 1.3 · sources: 2
Grok positions unchanged USA · Grok 4.7
ChatGPT Hold · 51% chance of a rise · Target: 66,752 · Wrong if: 63,000 USA · GPT-6.1 Sol · sources: 2
Nemotron positions unchanged USA · Nemotron 3 Ultra
Amazon Nova Buy on dips · 60% chance of a rise · Target: 70,000 · Wrong if: 62,000 · Entry: 64,000 USA · Amazon Nova · links opened: 0 of 1 · sources: 1Kimi† answer 2:48 PM ET Trim · 40% chance of a rise · Target: 65,358 · Wrong if: 68,500 China · Kimi K3 · sources: 2 · 1 confirm the figureQwen† answer 2:47 PM ET Trim · 35% chance of a rise · Target: 70,000 · Wrong if: 62,000 China · Qwen3.8-Max · sources: 1MiMo (Xiaomi)† answer 2:48 PM ET Buy on dips · 52% chance of a rise · Target: 67,500 · Wrong if: 64,000 · Entry: 64,500 China · MiMo-V2.6-Pro · sources: 4 · 2 confirm the figureERNIE (Wenxin)† answer 7:17 PM ET Sell · 33% chance of a rise · Target: 63,100 · Wrong if: 67,800 China · ERNIE 5.1 · links opened: 0 of 1 · sources: 1DeepSeek Trim · 42% chance of a rise · Target: 65,500 · Wrong if: 69,500 China · DeepSeek V4.1 · sources: 1Z.ai† answer 2:50 PM ET Sell · 38% chance of a rise · Target: 65,000 · Wrong if: 69,500 China · GLM-5.3 · sources: 2 · 1 confirm the figureStepFun† answer 12:56 PM ET Hold · 40% chance of a rise · Target: 65,000 · Wrong if: 68,500 China · Step 5 Preview · sources: 2
† — this answer is scored from the price at the moment it was given (the time and price beside the model), to the same close as every answer, in the day’s shared score.
Rating
collecting history: 30 of 30 resolved forecasts · trading days: 10/10 Full rating →
The rule of the game: to tell 55% right calls from a coin takes about 145 sessions ≈ 7 months; the ranking comes earlier and sharpens every day. (with the link between the forecasts of one day measured in the book, ρ≈0.11)
Coverage and Brier for each AI ▾
In each cell: how many of the questions asked the AI answered with a probability; then its Brier (0 is perfect, 0.25 is a coin) on its own answers, and with every gap counted as a 50% answer. A 50% the AI said itself and a gap counted as 50% are told apart.
Who
Day
Week
Month
6 months
ChatGPT
answered 95% Brier 0.248 on its own answers, 0.248 with the gaps counted as 50%
answered 95% Brier 0.250 on its own answers, 0.250 with the gaps counted as 50% · said “50%” itself: 1
answered 95% no outcome yet — the horizon has not ended
answered 95% no outcome yet — the horizon has not ended
Claude
answered 100% Brier 0.247 on its own answers, 0.247 with the gaps counted as 50% · said “50%” itself: 33
answered 100% Brier 0.259 on its own answers, 0.258 with the gaps counted as 50% · said “50%” itself: 28
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 8
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 3
Gemini
answered 95% Brier 0.249 on its own answers, 0.249 with the gaps counted as 50% · said “50%” itself: 1
answered 95% Brier 0.255 on its own answers, 0.255 with the gaps counted as 50%
answered 95% no outcome yet — the horizon has not ended · said “50%” itself: 2
answered 95% no outcome yet — the horizon has not ended
Grok
answered 100% Brier 0.242 on its own answers, 0.242 with the gaps counted as 50% · said “50%” itself: 7
answered 100% Brier 0.259 on its own answers, 0.259 with the gaps counted as 50% · said “50%” itself: 7
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 20
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 12
Meta AI
answered 99% Brier 0.240 on its own answers, 0.240 with the gaps counted as 50% · said “50%” itself: 2
answered 99% Brier 0.255 on its own answers, 0.255 with the gaps counted as 50% · said “50%” itself: 4
answered 99% no outcome yet — the horizon has not ended · said “50%” itself: 3
answered 99% no outcome yet — the horizon has not ended · said “50%” itself: 1
Nemotron
answered 85% Brier 0.258 on its own answers, 0.258 with the gaps counted as 50% · said “50%” itself: 8 · gaps counted as 50%: 9
answered 85% Brier 0.254 on its own answers, 0.253 with the gaps counted as 50% · said “50%” itself: 12 · gaps counted as 50%: 4
answered 85% no outcome yet — the horizon has not ended · said “50%” itself: 18
answered 85% no outcome yet — the horizon has not ended · said “50%” itself: 6
Amazon Nova
answered 83% Brier 0.250 on its own answers, 0.250 with the gaps counted as 50% · said “50%” itself: 18
answered 83% Brier 0.307 on its own answers, 0.300 with the gaps counted as 50% · said “50%” itself: 19
answered 83% no outcome yet — the horizon has not ended · said “50%” itself: 44
answered 83% no outcome yet — the horizon has not ended · said “50%” itself: 26
DeepSeek
answered 95% Brier 0.244 on its own answers, 0.244 with the gaps counted as 50% · said “50%” itself: 15
answered 95% Brier 0.258 on its own answers, 0.258 with the gaps counted as 50% · said “50%” itself: 11
answered 95% no outcome yet — the horizon has not ended · said “50%” itself: 12
answered 95% no outcome yet — the horizon has not ended · said “50%” itself: 6
Qwen
answered 89% Brier 0.237 on its own answers, 0.239 with the gaps counted as 50% · said “50%” itself: 23
answered 89% Brier 0.288 on its own answers, 0.286 with the gaps counted as 50% · said “50%” itself: 7
answered 89% no outcome yet — the horizon has not ended · said “50%” itself: 14
answered 89% no outcome yet — the horizon has not ended · said “50%” itself: 14
Kimi
answered 90% Brier 0.252 on its own answers, 0.252 with the gaps counted as 50% · said “50%” itself: 10
answered 90% Brier 0.269 on its own answers, 0.268 with the gaps counted as 50% · said “50%” itself: 9
answered 90% no outcome yet — the horizon has not ended · said “50%” itself: 4
answered 90% no outcome yet — the horizon has not ended · said “50%” itself: 11
Z.ai
answered 100% Brier 0.243 on its own answers, 0.244 with the gaps counted as 50% · said “50%” itself: 6
answered 100% Brier 0.269 on its own answers, 0.268 with the gaps counted as 50% · said “50%” itself: 9
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 7
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 10
ERNIE (Wenxin)
answered 84% Brier 0.249 on its own answers, 0.249 with the gaps counted as 50% · said “50%” itself: 21
answered 84% Brier 0.243 on its own answers, 0.244 with the gaps counted as 50% · said “50%” itself: 9
answered 84% no outcome yet — the horizon has not ended · said “50%” itself: 11
answered 84% no outcome yet — the horizon has not ended · said “50%” itself: 10
StepFun
answered 95% Brier 0.244 on its own answers, 0.245 with the gaps counted as 50% · said “50%” itself: 33
answered 95% Brier 0.254 on its own answers, 0.254 with the gaps counted as 50% · said “50%” itself: 17
answered 95% no outcome yet — the horizon has not ended · said “50%” itself: 8
answered 95% no outcome yet — the horizon has not ended · said “50%” itself: 26
MiMo (Xiaomi)
answered 100% Brier 0.252 on its own answers, 0.252 with the gaps counted as 50% · said “50%” itself: 8
answered 100% Brier 0.246 on its own answers, 0.247 with the gaps counted as 50% · said “50%” itself: 15
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 8
answered 100% no outcome yet — the horizon has not ended · said “50%” itself: 3
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.
One reading for the 10-year yield and VIX: the day’s call is a bet on the indicator itself (buy = the yield or the VIX rises); in the portfolio leagues the 10-year is the IEF fund, held long, and it rises when the yield falls.
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 before it starts to count, two leagues — long-only and Pro with 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 , 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 — 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 the mode named in its Configuration line. 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.
Decisions any day by the published rule (arena-protocol.json): the weekly review, the radar and the league guardian — a model with no accepted decision this week is asked again hourly 07:00–21:00 ET, at most 5 times a day; every decision trades at the next open; the Council’s position changes only in the scheduled issue. Counted in the leagues (arena-decisions.csv): the league guardian — main league 1, Pro league 2; the radar — main league 4, Pro league 2.
The league's protocol: version 3 and its history
version 1 (Sun, Oct 18, 2026): the first rules: long only, 5–25% a position, the rest cash at the T-bill, 0.1% on what changes, the first price of 09:30–09:45 ET.
version 2 (Mon, Oct 19, 2026): “hold” is a decision with no trade and no cost; each model sees only its own account; VIX became an indicator only; a decision any day.
version 3 (Tue, Oct 20, 2026): a fund may name its market (SPY = S&P 500); the model states its mandate; the positions and cash must sum to 100 within 0.5. The same weights again bring back a position that drifted 2 p.p. or more.
The day 2026-11-13: best — ChatGPT +$11 (+0.01%) · worst — Amazon Nova −$1,168 (−1.22%)
ChatGPT made the most on Dow Jones (+$26) · Amazon Nova lost the most on Chip index (−$365) · They bet apart on Nasdaq-100: Amazon Nova held 21%, ChatGPT none
Overtakes of the day: Z.ai passed MiMo (Xiaomi); Gemini passed MiMo (Xiaomi); StepFun passed MiMo (Xiaomi); Kimi passed DeepSeek; Meta AI passed Qwen
The league. Session Fri, Nov 13, 2026: traded — MiMo (Xiaomi), Claude, DeepSeek, Qwen, StepFun · answered and kept the weights — ChatGPT, Z.ai, Gemini, Kimi, Grok, Meta AI · on the bench, positions carried — ERNIE (Wenxin) · positions unchanged — Nemotron, Amazon Nova.
Participant
Value
Week
Return
vs S&P 500, p.p.
Max drawdown
Return per risk
Cash, avg
Real diversification
Turnover
Last action
MiMo (Xiaomi)
$96,642
+$969 +1.01%
−3.36%
−1.53
−4.73%
−2.50
30%
6.1
1.68
traded: 7 trades Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Nemotron
$96,273
+$215 +0.22%
−3.73%
−1.90
−4.91%
−3.30
24%
7.9
1.80
holds its positions Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
StepFun
$96,952
−$469 −0.48%
−3.05%
−1.22
−3.29%
−3.32
29%
4.2
1.82
traded: 6 trades Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
ChatGPT
$98,209
−$358 −0.36%
−1.79%
+0.03
−2.28%
−3.53
34%
4.2
1.11
held on purpose Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Claude
$96,132
−$287 −0.30%
−3.87%
−2.04
−4.41%
−4.45
22%
4.9
1.35
traded: 7 trades Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
ERNIE (Wenxin)
$94,686
−$1,251 −1.30%
−5.31%
−3.49
−7.56%
−4.53
21%
5.4
1.06
held on purpose Thu, Nov 12, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Gemini
$97,115
−$628 −0.64%
−2.88%
−1.06
−3.67%
−4.71
21%
4.6
0.94
held on purpose Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
DeepSeek
$95,573
−$828 −0.86%
−4.43%
−2.60
−4.87%
−4.72
14%
5.1
1.37
traded: 7 trades Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Grok
$95,433
−$1,243 −1.29%
−4.57%
−2.74
−4.97%
−4.88
27%
5.0
0.93
held on purpose Fri, Nov 13, 2026 decisions: 4 of 4 weeks
Amazon Nova
$94,705
−$833 −0.87%
−5.30%
−3.47
−5.83%
−5.08
22%
6.7
1.61
holds its positions Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Kimi
$95,833
−$697 −0.72%
−4.17%
−2.34
−4.88%
−5.45
23%
3.9
2.47
held on purpose Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Meta AI
$94,356
−$368 −0.39%
−5.64%
−3.82
−6.65%
−5.51
15%
5.3
3.67
held on purpose Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Qwen
$94,172
−$1,373 −1.44%
−5.83%
−4.01
−6.84%
−6.07
18%
6.1
4.32
traded: 8 trades Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Z.ai
$97,271
−$601 −0.61%
−2.73%
−0.91
−3.35%
−6.83
53%
4.9
1.20
held on purpose Fri, Nov 13, 2026 new decision accepted, takes effect Mon, Nov 16, 2026 decisions: 4 of 4 weeks
Yardsticks to compare with
S&P 500, bought once and held · from the 10-19 open, first price 09:30–09:45 ET, after costs
$98,177
+$2,483 +2.59%
−1.82%
—
−5.19%
−1.11
0%
1.0
1.00
held on purpose Mon, Nov 9, 2026 decisions: 4 of 4 weeks
60/40 stocks/bonds (outside yardstick, above the 25% cap)
$98,910
+$737 +0.75%
−1.09%
+0.73
−2.52%
−1.29
0%
1.9
1.03
traded: 2 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
The Council by confidence (each weight times the AI's own confidence)
$99,053
+$28 +0.03%
−0.95%
+0.88
−1.12%
−2.73
71%
3.0
0.65
traded: 4 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
The Council with its disagreement (more disagreement — more cash)
$98,331
−$199 −0.20%
−1.67%
+0.15
−1.96%
−3.17
59%
3.8
0.76
traded: 4 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
Where 3+ models agree (those markets alike)
$94,381
+$1,306 +1.40%
−5.62%
−3.80
−8.53%
−3.18
0%
6.0
3.41
traded: 7 trades Fri, Nov 13, 2026 decisions: 4 of 4 weeks
Equal weight
$94,296
−$377 −0.40%
−5.70%
−3.88
−6.96%
−4.75
0%
11.0
1.08
traded: 11 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
All traded markets, bought once and held
$94,171
−$501 −0.53%
−5.83%
−4.01
−7.02%
−4.99
0%
11.0
1.00
held on purpose Mon, Nov 9, 2026 decisions: 4 of 4 weeks
Random within the rules
$92,669
−$1,482 −1.57%
−7.33%
−5.51
−8.12%
−5.46
8%
5.8
4.25
traded: 10 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
Where 1–2 models are (the rest alike)
$93,101
−$584 −0.62%
−6.90%
−5.08
−8.24%
−5.61
2%
4.0
11.22
traded: 5 trades Fri, Nov 13, 2026 decisions: 4 of 4 weeks
Team USA (the US models' mean weights)
$96,929
−$581 −0.60%
−3.07%
−1.25
−3.38%
−6.02
34%
5.9
1.31
traded: 9 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
Team China (the Chinese models' mean weights)
$95,828
−$728 −0.75%
−4.17%
−2.35
−4.58%
−6.49
37%
3.8
1.71
traded: 7 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
The Council by past accuracy (each AI weighted by its right calls)
$95,702
−$651 −0.68%
−4.30%
−2.48
−4.65%
−6.91
37%
4.7
1.33
traded: 8 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
The Council (mean of the AIs' weights)
$96,398
−$736 −0.76%
−3.60%
−1.78
−3.95%
−7.10
42%
3.8
1.35
traded: 7 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
The Council by majority (markets most AIs hold, alike)
$93,068
−$1,730 −1.82%
−6.93%
−5.11
−7.64%
−7.42
0%
4.0
2.37
traded: 5 trades Mon, Nov 9, 2026 decisions: 4 of 4 weeks
Cash only
$100,281
+$79 +0.08%
+0.28%
+2.10
0.00%
—
100%
—
0.00
kept the weights Mon, Nov 9, 2026 decisions: 4 of 4 weeks
The readers (the median of the readers' own portfolios, under the same rules)
$100,281
+$79 +0.08%
+0.28%
+2.10
0.00%
—
100%
—
0.00
positions unchanged decisions: 0 of 4 weeks
The market over the same period: S&P 500 −1.82%, Bitcoin −16.51%, “equal” −5.70%. The market fell. The whole field's day return moves with the S&P 500 at a correlation of 0.88 — a lean, not skill.
“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 (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.
Overtakes: Fri, Nov 13, 2026 Z.ai passed MiMo (Xiaomi); Fri, Nov 13, 2026 Gemini passed MiMo (Xiaomi); Fri, Nov 13, 2026 StepFun passed MiMo (Xiaomi); Fri, Nov 13, 2026 Kimi passed DeepSeek; Fri, Nov 13, 2026 Meta AI passed Qwen; Thu, Nov 12, 2026 Z.ai passed Gemini; Thu, Nov 12, 2026 Z.ai passed StepFun; Thu, Nov 12, 2026 Gemini passed StepFun
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.
Right more often
#
Participant
Right
Calls
Place by money minus place here
1
Qwen
60%
143
+13
2
Z.ai
59%
192
+0
3
Amazon Nova
57%
129
+8
4
Grok
57%
215
+6
5
Meta AI
57%
200
+8
6
Claude
55%
190
+1
7
DeepSeek
55%
185
+2
8
StepFun
54%
158
−4
9
Gemini
53%
207
−6
10
ChatGPT
52%
196
−9
11
ERNIE (Wenxin)
52%
131
+1
12
MiMo (Xiaomi)
52%
212
−7
13
Kimi
52%
176
−5
14
Nemotron
51%
150
−8
Earned more
#
Participant
Return
1
ChatGPT
−1.79%
2
Z.ai
−2.73%
3
Gemini
−2.88%
4
StepFun
−3.05%
5
MiMo (Xiaomi)
−3.36%
6
Nemotron
−3.73%
7
Claude
−3.87%
8
Kimi
−4.17%
9
DeepSeek
−4.43%
10
Grok
−4.57%
11
Amazon Nova
−5.30%
12
ERNIE (Wenxin)
−5.31%
13
Meta AI
−5.64%
14
Qwen
−5.83%
The Council −3.60% against the mean model −4.05% and the best one, ChatGPT, −1.79% — the same period.
Where 3+ models agreed at the last decision: Bitcoin — 0 raised, 3 cut, Copper — 0 raised, 3 cut.
The week and the season in a few words
The week since 2026-11-09: best MiMo (Xiaomi) +1.01%, worst Qwen −1.44%; “equal” −0.40%, S&P 500 +2.59%; all the AIs' costs this week $392. MiMo (Xiaomi) made the most on Chip index (+$1,400). Since the season began: ahead ChatGPT −1.79% (deepest fall 2.28%), behind Qwen −5.83%; “equal” −5.70%, S&P 500 −1.82%. A story, not a verdict.
Your own virtual $100,000 — against the AIs
The same rules: each position 5–25%, the rest in cash. Your weights are sealed by the server's clock and trade at the next open, like the AIs'.
All accounts on one chart
MiMo (Xiaomi)NemotronStepFunChatGPTClaudeERNIE (Wenxin)GeminiDeepSeekGrokAmazon NovaKimiMeta AIQwenZ.aiS&P 500, bought once and held60/40 stocks/bonds (outside yardstick, above the 25% cap)The Council by confidence (each weight times the AI's own confidence)The Council with its disagreement (more disagreement — more cash)Where 3+ models agree (those markets alike)Equal weightAll traded markets, bought once and heldRandom within the rulesWhere 1–2 models are (the rest alike)Team USA (the US models' mean weights)Team China (the Chinese models' mean weights)The Council by past accuracy (each AI weighted by its right calls)The Council (mean of the AIs' weights)The Council by majority (markets most AIs hold, alike)Cash onlyThe readers (the median of the readers' own portfolios, under the same rules)
Season 1 (seasons of 126 sessions are slices of one continuous account), from 2026-10-19: session 20.
Beat the simple rules after risk: MiMo (Xiaomi), Nemotron, StepFun, ChatGPT, Claude, ERNIE (Wenxin), Gemini, DeepSeek
This week's weights
MiMo (Xiaomi): S&P 500 15%, Dow Jones 10% → 13%, Copper 14% → 13%, Chip index 23% → 10% · new decision accepted, takes effect Mon, Nov 16, 2026 day −$899 (−0.92%) · week $969 (+1.01%) · month −$294 (−0.30%) ⚠ Trades against its own forecast: 2. Latest: Chip index +8.0 p.p. while it gave 45% for a rise this week.Why and the mix
Nemotron: S&P 500 18%, Dow Jones 12%, US 10-year bonds (IEF) 10% → 12%, Gold 7% → 10% · holds its positions · new decision accepted, takes effect Mon, Nov 16, 2026 day −$846 (−0.87%) · week $215 (+0.22%) · month −$2,047 (−2.08%) ⚠ Trades against its own forecast: 2. Latest: Bitcoin +5.0 p.p. while it gave 42% for a rise this week.Why and the mix
StepFun: Dow Jones 20% → 22%, Dollar index 12% → 15%, Copper 8% → 9%, Gold 11% → 8% · Bitcoin 5% → 0, Nasdaq-100 5% → 0 · new decision accepted, takes effect Mon, Nov 16, 2026 day −$280 (−0.29%) · week −$469 (−0.48%) · month −$957 (−0.98%) ⚠ Trades against its own forecast: 6. Latest: Gold +2.0 p.p. while it gave 45% for a rise this week.Why and the mix
ChatGPT: Dow Jones 20%, Gold 15%, S&P 500 5% → 10%, US 10-year bonds (IEF) 10% · Brent crude oil 5% → 0 · holds its positions · new decision accepted, takes effect Mon, Nov 16, 2026 day $11 (+0.01%) · week −$358 (−0.36%) · month −$576 (−0.58%) ⚠ Trades against its own forecast: 4. Latest: S&P 500 +5.0 p.p. while it gave 45% for a rise this week.Why and the mix
Claude: US 10-year bonds (IEF) 20%, S&P 500 12% → 17%, Dow Jones 10% → 13%, Nasdaq-100 8% · Bitcoin 5% → 0, Copper 5% → 0 · new decision accepted, takes effect Mon, Nov 16, 2026 day −$188 (−0.20%) · week −$287 (−0.30%) · month −$2,356 (−2.39%) ⚠ Trades against its own forecast: 3. Latest: S&P 500 −3.0 p.p. while it gave 56% for a rise this week.Why and the mix
ERNIE (Wenxin): Brent crude oil 20%, Dow Jones 20%, WTI crude oil 20% → 18%, Gold 0% → 12% · holds its positions · new decision accepted, takes effect Mon, Nov 16, 2026 day −$192 (−0.20%) · week −$1,251 (−1.30%) · month −$6,971 (−6.86%) ⚠ Trades against its own forecast: 3. Latest: Brent crude oil +2.0 p.p. while it gave 45% for a rise this week.Why and the mix
Meta AI: Gold 15% → 20%, Dollar index 20% → 19%, Dow Jones 15% → 18%, S&P 500 15% → 9% · Brent crude oil 5% → 0, WTI crude oil 5% → 0 · holds its positions · new decision accepted, takes effect Mon, Nov 16, 2026 day −$251 (−0.27%) · week −$368 (−0.39%) · month −$5,817 (−5.81%) ⚠ Trades against its own forecast: 2. Latest: S&P 500 +3.0 p.p. while it gave 42% for a rise this week.Why and the mix
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.
Participant
Equity
Week
Return
vs S&P 500, p.p.
Max drawdown
Invested
State
Gemini Since Gemini 4 Argon · thinking_level: high · Google Search grounding took over (2026-10-26): +0.88%. (with the positions it took over)
$102,475
+$547 +0.54%
+2.47%
+4.03
−1.08%
139%
—
StepFun
$99,887
−$3,035 −2.95%
−0.11%
+1.44
−4.85%
112%
—
MiMo (Xiaomi)
$97,992
−$4,792 −4.66%
−2.01%
−0.45
−5.77%
111%
—
Z.ai Since GLM-5.3 · thinking: enabled · web search took over (2026-11-11): −0.71%. (with the positions it took over)
$97,366
−$3,644 −3.61%
−2.63%
−1.08
−5.50%
104%
—
ChatGPT Since GPT-6.1 Sol · reasoning.effort: max · web search took over (2026-11-10): −1.04%. (with the positions it took over)
$97,313
−$3,245 −3.23%
−2.69%
−1.13
−3.99%
100%
—
ERNIE (Wenxin) Since 文心 5.1 · web search took over (2026-11-02): −7.89%. (with the positions it took over)
$96,464
−$3,884 −3.87%
−3.54%
−1.98
−9.96%
102%
—
DeepSeek
$96,324
−$4,887 −4.83%
−3.68%
−2.12
−5.38%
133%
—
Nemotron
$96,122
−$363 −0.38%
−3.88%
−2.32
−5.11%
101%
—
Grok
$95,756
−$1,514 −1.56%
−4.24%
−2.69
−4.84%
111%
—
Qwen Since Qwen3.8-Max · Max + search took over (2026-10-26): −3.68%. (with the positions it took over)
$94,365
+$3,289 +3.61%
−5.64%
−4.08
−10.57%
109%
—
Claude
$94,334
+$19 +0.02%
−5.67%
−4.11
−6.46%
101%
—
Amazon Nova
$94,319
−$243 −0.26%
−5.68%
−4.12
−6.65%
109%
—
Meta AI
$94,117
−$3,819 −3.90%
−5.88%
−4.33
−10.85%
181%
—
Kimi
$92,068
−$519 −0.56%
−7.93%
−6.38
−9.54%
111%
—
Yardsticks to compare with
S&P 500, bought once and held since Tue, Oct 20, 2026 · from the 10-20 open, first price 09:30–09:45 ET, after costs
$98,444
+$2,490 +2.59%
−1.56%
—
−5.19%
100%
—
Always the bear: the S&P 500 short with the whole capital
$102,116
−$2,343 −2.24%
+2.12%
+3.67
−3.22%
96%
—
The day 2026-11-13: best — Meta AI +$1,483 (+1.60%) · worst — Nemotron −$812 (−0.84%)
Overtakes: Fri, Nov 13, 2026 Z.ai passed ChatGPT; Fri, Nov 13, 2026 Z.ai passed DeepSeek; Fri, Nov 13, 2026 Z.ai passed Nemotron; Fri, Nov 13, 2026 ERNIE (Wenxin) passed DeepSeek; Fri, Nov 13, 2026 ERNIE (Wenxin) passed Nemotron; Fri, Nov 13, 2026 DeepSeek passed Nemotron; Fri, Nov 13, 2026 Qwen passed Claude; Fri, Nov 13, 2026 Qwen passed Amazon Nova
2026-11-13 — Qwen reversed S&P 500 at a loss of −$54 Why: Flipping tech and broad equities to short as yields break out, while going long VIX to capture the inevitable volatility spike from a Fed policy error.
2026-11-13 — Claude closed Copper at a loss of −$307 Why: Copper broke the 6.20 line I use in both books, so I sell the pro copper (4.9%) and move it into the dollar (10% to 12%), the one position that gains if PPI and the Fed speeches turn more hawkish. The VIX short is squared to 5%, and everything else stays as it is (gross 101%).
2026-11-13 — Amazon Nova closed Dollar index at a loss of −$171 Why: Aggressive long commodities, tech; short rates and SOX for leverage.
2026-11-11 — Qwen reversed Dollar index at a loss of −$102 Why: Flipping US10Y to short as December hike odds surge to 88%, and flipping DXY to long on rate differentials.
2026-11-11 — Qwen reversed US 10-year bonds (IEF) at a loss of −$166 Why: Flipping US10Y to short as December hike odds surge to 88%, and flipping DXY to long on rate differentials.
2026-11-10 — StepFun closed WTI crude oil at a loss of −$2,916 Why: Brent's break below $90 is my trigger: I close the WTI long, open a 10% Brent short on the IEA glut, trim the dollar to 16% and VIX to 8%, and add to the Dow to 28%; the NDX, SOX, bond and BTC shorts stay.
2026-11-10 — Nemotron closed Nasdaq-100 at a loss of −$189 Why: Long China stimulus + Fed pivot basket (SOX, Copper, Gold, SPX, DJI). Short US10Y yield (long IEF) and DXY on rate differential compression. Long VIX for CPI/Fed event vol. Small BTC for optionality. Gross 101%.
2026-11-10 — Nemotron reversed Fear index (VIX) at a loss of −$165 Why: Long China stimulus + Fed pivot basket (SOX, Copper, Gold, SPX, DJI). Short US10Y yield (long IEF) and DXY on rate differential compression. Long VIX for CPI/Fed event vol. Small BTC for optionality. Gross 101%.
2026-11-10 — Nemotron closed Brent crude oil at a loss of −$868 Why: Long China stimulus + Fed pivot basket (SOX, Copper, Gold, SPX, DJI). Short US10Y yield (long IEF) and DXY on rate differential compression. Long VIX for CPI/Fed event vol. Small BTC for optionality. Gross 101%.
2026-11-10 — Amazon Nova closed Gold at a loss of −$213 Why: Long tech, crypto, volatility; short rates and dollar for leverage.
2026-11-10 — Meta AI reversed Copper at a loss of −$9 Why: I keep the hawkish Fed, AI unwind and peace-oil short, flip copper long on China imports, and stay long dollar, gold and vol as hedges.
2026-11-10 — Kimi closed Fear index (VIX) at a loss of −$552 Why: I close the VIX short into the 8:30 CPI print — a hot number spikes vol straight against it — and keep the barbell: long Dow, gold, SPX, copper and the dollar against short oil, BTC, NDX, SOX and the 10-year bond.
2026-11-13 — StepFun closed Bitcoin at a loss of −$845 Why: Sell BTC (the rate-shock loser, now under the 5% floor), trim gold and copper into strength, add DXY — the cleanest December-hike expression — and keep about 38% cash for the Dec 9/10 one-two punch.
2026-11-13 — Claude closed Copper at a loss of −$311 Why: My copper stop fired (COMEX 6.15, under 6.20, after LME's four-month low), so the 5% copper goes to cash, now 30%. No other rule fired: the S&P is at 7,592 vs my 7,300 line, the 10-year at 5.10% vs 5.20%, gold at 3,893 vs 3,850, and there is no Hormuz deal.
2026-11-13 — DeepSeek closed Bitcoin at a loss of −$847 Why: I sell the BTC stub after the 5% break and keep cash near 28% ahead of the December hike.
2026-11-13 — Qwen closed WTI crude oil at a loss of −$323 Why: Trimming tech exposure and raising cash to prepare for a volatility spike, while increasing long dollar and value equity positions to hedge against sticky inflation.
2026-11-11 — Kimi closed Brent crude oil at a loss of −$728 Why: Hot CPI and Waller's hawkish shift are important changes: I exit the small Brent and IEF positions, raise cash to 35%, add a 5% dollar position, and keep Dow, SPX, gold and copper.
2026-11-10 — StepFun closed Nasdaq-100 at a loss of −$324 Then, 2026-11-13: bought Dollar index, S&P 500. Why: Brent broke below $90 overnight on Iran de-escalation — my trigger level — so I cut the oil-linked dollar from 12% to 8%, closed the Nasdaq stub (bearish into the December hike) and raised cash into today's 08:30 CPI; I add to the Dow on the value rotation and to copper on China's 10-trillion-yuan d
2026-11-10 — Meta AI closed Brent crude oil at a loss of −$931 Why: I sell all oil on unconfirmed Iran peace plus Ukraine talks, rotate to Dow, dollar and gold, and add copper on strong China imports.
2026-11-10 — Meta AI closed WTI crude oil at a loss of −$783 Why: I sell all oil on unconfirmed Iran peace plus Ukraine talks, rotate to Dow, dollar and gold, and add copper on strong China imports.
2026-11-09 — ChatGPT closed Brent crude oil at a loss of −$509 Why: At $98,533, down 1.47% from inception with a maximum drawdown of 2.00%, I prioritize capital preservation while retaining recovery exposure. I exit the roughly 5% Brent position because its geopolitical premium is contracting, increase broad-market exposure toward 10%, and retain the Dow, gold, IEF
2026-11-09 — Claude closed Bitcoin at a loss of −$487 Then, 2026-11-12: bought S&P 500, US 10-year bonds (IEF). Why: I sell bitcoin (down 1.8% on a risk-on morning, $870M of ETF outflows in four days, sitting on my 74,000 line), cut the Brent hedge from 7% to 5% after its steepest weekly fall since 2020, and lift gold to 7% with central-bank buying at a 2026 high; stocks stay at 30% until CPI and Nvidia are out.
2026-11-09 — Gemini closed Brent crude oil at a loss of −$1,030 Why: At this weekly review I am liquidating our 9.6% Brent position—which is sliding 3.27% today as the crude curve rolls over on a 903,000 b/d OPEC+ supply surge and falling Chinese oil demand—and rotating those proceeds into Gold (18%), the Dow (22%), and a 25% cash buffer while maintaining zero exposu
2026-11-09 — Amazon Nova closed Brent crude oil at a loss of −$1,553 Why: Reduce exposure to rate sensitivity, increase tech and commodities; maintain liquidity for dips.
Every simulated trade of the virtual league (last 40)
Stated confidence against position size: correlation 0.53 over 654 positions.
Virtual $100,000 portfolios by actions — a separate count
Each AI trades its own $100,000 with its actions, after costs. Sessions so far: 20. 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.
As of Nov 13 · sessions so far: 20 · the AIs after costs, the S&P 500 without costs
Who
Value
Return
In the market
StepFun
$100,351
▲ +0.35%
35%
DeepSeek
$99,816
▼ −0.18%
80%
Z.ai
$98,831
▼ −1.17%
51%
ChatGPT
$98,830
▼ −1.17%
35%
Kimi
$98,411
▼ −1.59%
20%
MiMo (Xiaomi)
$98,282
▼ −1.72%
10%
S&P 500, buy and hold · since 10-19 · from the 10-19 open, first price 09:30–09:45 ET, without costs
$98,275
▼ −1.72%
100%
The Council by actions
$97,908
▼ −2.09%
70%
Nemotron
$97,649
▼ −2.35%
100%
Claude
$96,992
▼ −3.01%
40%
Gemini
$96,844
▼ −3.16%
100%
Grok
$96,604
▼ −3.40%
31%
Meta AI
$95,921
▼ −4.08%
41%
Qwen
$94,561
▼ −5.44%
86%
Amazon Nova
$93,554
▼ −6.45%
100%
ERNIE (Wenxin)
$93,034
▼ −6.97%
95%
Calibration: how often the stated odds came true
When the AIs said “60–70% for a rise”, how often it really happened.
They said
On average
Came true
Forecasts
[50,60)
54%
53% n=1911 · 95% interval 51–55%
1911
[60,70)
62%
59% n=491 · 95% interval 54–63%
491
[70,80)
71%
73% n=79 · 95% interval 63–82%
79
[80,100]
80%
67% n=3 · 95% interval 21–94%
3
Bins: [50,60) [60,70) [70,80) [80,100] — the lower edge in, the upper out, 100 in the last; a stated 50% backs no side and is in none.
Against the simple rules (, lower is better; coin 0.25): the AIs' 0.244 · weighted by past accuracy 0.244 · USA 0.245 · China 0.244 — over 214 indicator-sessions. Momentum (yesterday’s direction repeated, a 55% call): Brier 0.250 over 186 indicator-sessions.
The sites the AIs cite — how often their call with it came true (calls): home.saxo 87% (15) · bls.gov 71% (48) · stocktitan.net 71% (34) · benzinga.com 70% (20) · mitrade.com 69% (16)
The AIs against the crowd: the score
All days, the individual AIs’ answers: 36 of 51, the crowd 35 of 51 · Brier: the AIs 0.218, the crowd 0.176 (lower is better).
, all days: the AIs 8 of 9, the crowd 9 of 9 · Brier: the AIs 0.197, the crowd 0.071 (lower is better).
The models' own lines on the way: “target” reached 1758/3795, “wrong if” touched 1656/4717.
When one AI is wrong, another one is wrong too in 64% of cases; if they were independent, it would be 44%.
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.
Registered in advance on OSF (outcomes F28, F29) — osf.io/9q86n · target: 200 first calls in each group
Collecting calls: the share shows from 10 in each group.
Shadow modes: the same AIs under other conditions
Each mode has its own accounts and its own open file; the league’s score is counted without them.
As of Nov 13 · sessions so far: 19, from Oct 20 · the AIs after costs
12 funds — each AI’s own portfolio held in the 12 funds themselves (SOXX, TLT, IBIT and the others) at their own prices, rules f12-1. Best on Nov 13, 2026: ChatGPT: $98,322, −1.68%. · arena-accounts.csv · arena-accounts-dictionary.json
Closed book — after the issue the same question without web search, on one sealed packet of news. The same news — less dispute? Effective voices, closed minus open: −0.60 over 12 runs (fewer — the answers closer to one another). · arena-cb-experiment.json · arena-cb-packets.json · arena-cb-league.json
Repeats — once a week the same portfolio question to the same model again. One model’s weights move against themselves by 0.3 percentage points on average; its own noise is 23% of the heat of the dispute between the models. · council-repeats.json
Two routes — the same question, word for word, through the model’s website and its command line. The decision to hold differed in 7 of 35 pairs; in the other pairs the distance between the two portfolios is 0.02 (0 — the same portfolio). · arena-modalities.json
Roles — the same model asked to trade in a role of character or at a mode of risk. Each role’s account against the model’s own, in percentage points on average: from YOLO −5.52 to Steady Eddie +0.64. · arena-roles.json
S&P 100 — a second game: the stocks of the S&P 100, the composition and its sha256 in the file. Best after 19 sessions: ChatGPT: +3.51%. As of Nov 13 · the AIs after costs · arena-sp100-league.json · arena-sp100-trades.csv
The Council after the discussion — the models criticise each other’s portfolios and a chair seals one portfolio. Against the simple mean of the models since Oct 20, 2026: +0.06 percentage points. · council-talk.json
Schools of thought — the same model and question with an American and a Chinese analyst’s school: two accounts and the gap between them in money · starts Oct 26, 2026 when switched on
Every model in every role — each model takes the roles in turn, one role a day: the table model × role · starts Oct 26, 2026 when switched on
Up / draw / down — three chances summing to 100 and the Brier on the triple beside the plain up-or-down one · starts Oct 26, 2026 when switched on
Score of the day
The score comes after the close of the session: Nov 16, 2026, 4:30 PM
For pros
Everything here can be recounted by anyone: the method, the protocol (version 3), the sealed book, every forecast as a table, the code of the count. Free to use, CC BY 4.0.
/api/v1/ — all open files, their columns and how often they change
markets-digest.json — a short digest of this page for AI readers (under 30 KB)
Check it yourself in 3 steps
On the proof page open the stamped list <time>.txt: for daily/council/book.jsonl it gives the number of lines N and its sha256. Download council-book.jsonl and run head -n N council-book.jsonl | shasum -a 256 : the hashes match (the book only grows, so the hash of the whole file never would). Download the .txt beside its .ots too.
On the proof page take the day’s .ots file and run ots verify <file>.txt.ots (pip install opentimestamps-client): it names the Bitcoin block that holds the stamp.
Recount: the probability and move columns of council-forecasts.csv give the Brier score and the calibration (how often the 70% calls came true); the recount script recounts the league’s accounts.
Each model’s seat is a subscription, so the calls have a price. Spent: $26.46 today · $396.99 this month · $806.80 since 2026-10-16; 1165 calls, $0.69 a call on average. The prices are the providers’ list prices.
After costs the best is ChatGPT: −2.02% (before costs −1.79%).
S&P 500 up and VIX down — calm growth. Yields up and the dollar up — pressure on gold, oil and emerging markets. Gold and VIX up together — fear. Copper up — factories and construction are speeding up, often China. Bitcoin moves first at weekends, when the exchanges are closed.
The same files the paper uses, read on this build: what the numbers cover, how close the AIs are to the crowd and to each other, and how each figure is checked.
The Spread: where the AIs and the crowd differ most
The crowd’s own contracts with money traded, the widest gap first: the crowd’s price and the AIs’ mean chance of the same event, in percent; the gap is AIs minus crowd, in points.
Effective voices of this issue by horizon: Day 1.7 · Week 1.9 · Month 2.0 · 6 months 1.9.
Scored sessions: on the same side of 50%: 92% of 49 contracts · distance from 50: the crowd 22.7, the AIs 8.9 points. All contracts with money: 87% of 54.
Evidence of order for this issue (answers sealed before the reference price): level 2+ — two independent timestamp authorities signed the list before the reference price
Scored sessions: 20 · data as of 2026-11-15
Climatology (the 2016–2025 base rate) — the same event exactly: S&P 500, Nasdaq-100, Dow Jones, Chip index; the nearest daily bar: US 10-year, Dollar index, Fear index (VIX), Gold, WTI crude oil, Brent crude oil, Copper, Bitcoin
Portfolios: virtual $100,000 accounts, traded by fixed rules with costs
US and Chinese AIs: an observed comparison on the same questions at the same minute
Scientific checks of the figures
issue of 2026-11-14: Horizon profile: from a day to six months · How distinct each AI’s view is · Do US and Chinese AIs form two blocks? The test and the map …
A study of fine-tuned agents found their errors correlate at 0.70. Source How we differ: We do not borrow this figure: we measure the overlap between our own answers separately and show it on the page.
The PolyBench study, on live Polymarket data: 2 of 7 models ended in profit. Source How we differ: We use virtual money only and show every loss; our method differs, so the results are not directly comparable.
Polymarket lists about 600 markets on which model will lead LMArena. Source How we differ: For us this is only a possible source of questions, not a result and not proof of anyone's accuracy.
Prediction Arena, real-money trading: on Kalshi 6 models lost 22.6% on average, on Polymarket about 1.1%; Grok guessed 51.9% of markets. Source How we differ: We trade virtual money only, and we publish our losses too.
The Prophet Arena paper evaluated forecasters on 1,367 resolved real-world events. Source How we differ: We do not copy its setup; our questions, scoring and virtual money are our own, so the numbers are not comparable.
The Foresight Arena paper estimates that detecting a true edge of 0.02 at 80% power needs about 350 resolved binary predictions. Source How we differ: For us this is a reminder that small samples prove little; we show sample sizes next to results.
The FinBench pilot covered one trading day, three tickers and 33 forecasts. Source How we differ: A pilot this small shows little; we make no comparison with it.
A study of forecasting with retrieved online discussions tested on 216 resolved prediction markets. Source How we differ: We do not borrow its setup or numbers; we show our own results with sample sizes.
A paper on earnings-call mention markets evaluated 856 Kalshi markets. Source How we differ: Different markets and method; we only note that sample sizes there are large, and show ours next to results.
Who else does this, and how we differ
From our own research register. Outside projects are described only as the register describes them; where it says the substance is not checked, it is not checked.
Register item 71: Forecaster Arena, an open project, runs 7 models on Polymarket events; llm-predict-alpha works on short binary Polymarket markets. These are the closest comparisons; we study how they present results.
Register item 72: LightningRod offers a product for forecasting on prediction markets. Its substance is not checked.
Register item 73: Kalshi has an internal agent, Harrison, that reads news, watches competitors and proposes new contracts (Bloomberg, 15.06.2026). This matches our idea of a model proposing the day's questions.
Register item 77: Mantic raised $25 million (18.09.2026); a retelling says AI took 1st, 2nd and 5th place at the Metaculus Cup. Second-hand, dates differ between sources and are to be checked. It is an objection to our claims and we keep it in view.
Register item 78: Public launched agents for Kalshi (24.09.2026). Real money, not our concern; an agent acting on a probability trigger is an idea for a virtual league.
Register item 78: Polymarket closed its framework for agents (11.05.2026). Real money, not our concern.
Register item 45: Meta Arena, an internal Meta project (press reports, июнь 2026, second-hand): points only, no money, not released, no results. How we differ: ours is sealed, virtual $100,000, public.
Our difference: virtual money only, and the comparison is with the crowd's price on the same questions.
What speaks against us
Honest objections, each with what we check. Virtual money only.
In Alpha Arena Season 1.5 only 5 of 32 accounts ended in profit. We check: does our council stay ahead of the plain market baseline over many questions, not one good week.
In TradeRank season 2 only 3 of 13 agents made a profit, and all 3 were contrarian. We check: does a contrarian rule beat the crowd in our own sealed record.
Forecasting Research Institute: superforecasters score a Brier of 0.086, the best language models 0.103 (March 2026); parity is only an extrapolation. We compare with the crowd price, not with superforecasters, and we do not claim parity. We check the gap on our own sealed questions. ↗
The startup Mantic raised $25 million on 18 September 2026 after outscoring human entrants in a forecasting contest. Money raised is not proof of accuracy. We check every forecast with a seal made before the outcome.
Questions and answers
Who are these AIs?
Several independent AI models from different companies. Each gets the same question about every market, answers on its own, and every answer is shown under the model’s name.
Where does the data come from, and how can I check a forecast?
Prices are public market quotes. Every forecast is sealed with a timestamp and a hash before the New York open and counted against the real close afterwards. The sealed lists and all the data can be downloaded (CSV, JSON), so anyone can recount the score.
Are the AIs better than the market?
Not proven. So far 100 comparable questions are counted of the 350 needed for a statistical verdict; today the data do not show that the AIs beat the market.
Is this investment advice?
No. All money here is virtual. The forecasts are an experiment in how well AIs call markets, not a recommendation to buy or sell anything.
Models and modes:
In this issue (as sealed in council-forecasts.csv, run 2026-11-14-btc): GPT-6.1 Sol — reasoning.effort: xhigh · web search (the highest our account offers) · answered; Claude Opus 5.5 · answered; Gemini 4 Argon · answered; Grok 4.7 · answered; Muse Spark 1.3 · answered; Nemotron 3 Ultra + Mercury (search) · answered; Amazon Nova — maxReasoningEffort: high (the highest our account offers) · answered; DeepSeek V4.1 — thinking: enabled · reasoning_effort: high · web search (the highest our account offers) · answered; Qwen3.8-Max — Max + search · answered; Kimi K3 · answered; GLM-5.3 — thinking: enabled · web search (the highest our account offers) · answered; ERNIE 5.1 — web search (the highest our account offers) · answered; Step 5 Preview — reasoning_effort: high · web search (the highest our account offers) · answered; MiMo-V2.6-Pro — thinking: enabled · web search (the highest our account offers) · answered.
Configured flagship and mode (target); today’s actual model and mode are above. Each model is listed with the mode named next to it; the model and mode of each day’s answer are with the answer in the book: GPT-6.1 Sol — reasoning.effort: xhigh · web search; Claude Opus 5.5 — effort: max · web search; Gemini 4 Argon — thinking_level: high · Google Search grounding; Grok 4.7 — reasoning_effort: xhigh · web search; Muse Spark 1.3 — reasoning: on · search grounding; Nemotron 3 Ultra — enable_thinking: true · effort: high; Amazon Nova — maxReasoningEffort: high; DeepSeek V4.1 — thinking: enabled · reasoning_effort: high · web search; Qwen3.8-Max — enable_thinking: true · web search; Kimi K3 — reasoning_effort: max · web search; GLM-5.3 — thinking: enabled · web search; ERNIE 5.1 — web search; Step 5 Preview — reasoning_effort: high · web search; MiMo-V2.6-Pro — thinking: enabled · web search.
Prices: Yahoo Finance (exchange data, delay about 15 min), Bitcoin — Kraken. One “now” number on the site and in the app.
No sign-up, no wallet, no money: you read and make your calls.
Seal: The answers were sealed before the open (RFC 3161), then anchored in Bitcoin (OpenTimestamps): SHA-256 a6fa58501d97926a…. Verify the signature
An AI experiment — the same issue for every reader, not personal investment advice. ‡ These results are based on simulated or hypothetical performance results that have certain inherent limitations. Unlike the results shown in an actual performance record, these results do not represent actual trading. Also, because these trades have not actually been executed, these results may have under-or over-compensated for the impact, if any, of certain market factors, such as lack of liquidity. Simulated or hypothetical trading programs in general are also subject to the fact that they are designed with the benefit of hindsight. No representation is being made that any account will or is likely to achieve profits or losses similar to these being shown.Terms · Method and conflicts →
Page version 9202d5d57a02 · built 2026-11-15 08:54 UTC
Tail league · $10,000 virtual
Each day the cheap side (under 20 cents) of every crowd market gets a virtual $100. Contracts played: 42; paid: 2; went to zero: 40; still open: 0.
The $10,000 fund is now worth $6,929 (−$3,071 against the start). Staked in all: $4,200.
A win pays the stake divided by the price, so a few wins can outweigh many zeros. Every zero is listed below.
Small sample: 42 settled contracts. The table shows what cheap contracts cost, not a method that works.