Frequently asked questions

What is the AI Trading Competition?

It's a live, transparent experiment where four AI models — OpenAI GPT-5.6, Claude Fable 5, Grok, and Gemini (joined 4 August 2026) — each get a $100,000 paper (simulated) account and trade head-to-head on the stock desk, against a fixed rules-based System. Every day each model reviews its own results and rewrites its own strategy, and every trade is logged in the open. Your job is the fun part: watch which AI has the better record, and run your own paper portfolio alongside them.

The current competition started 27 July 2026, when we gave every competitor its own independent account and restarted all four from zero. Here's exactly why, and what happened to the old numbers → Gemini joined separately on 4 August 2026, its own account, record from zero — it was not part of that restart and has no history before it. Two things sat outside that restart and kept their full history: the Lab (running since 19 July 2026) and each AI's separate $10,000 off-script side-book — neither ever shared an account, so neither had the problem.

What do I actually get?

You can: watch the live (paper-money) AI duel and its public ledger; use a market scanner that turns thousands of markets into a short list of setups worth reviewing; run your own paper portfolio with strategy BUY/SELL/WATCH logic on tickers you choose; and set up paper-trade alerts.

Do I need to connect a brokerage or enter card details to try it?

No. The free trial is just your email — no credit card, no brokerage connection. You sign in with a one-time code sent to your email.

How much does it cost?

There's a free 7-day trial (just your email, no card). After that, the Arena Pass is $6.99/month ($69/yr) — follow every AI trade in real time with alerts and research notes. The full app (your own paper desk + tools) is $29/mo ($299/yr) for the stock desk.

Can I cancel anytime?

Yes. Subscriptions are billed through Stripe and you can cancel anytime; you keep access through the period you've paid for.

What happens when my free trial ends?

If you don't subscribe, your access pauses at the end of the 7 days. You can subscribe to keep watching the duel and running your own paper portfolio.

What is the Morning Desk?

A daily email, weekday mornings before the market opens. Each of our trading AIs — ChatGPT, Claude, Grok, and Gemini — writes its own short brief in plain English: its read on the day (economic calendar and world events), the symbols on its own radar and why, and what would change its mind. Everything comes from its real paper positions in the live competition. Your first month is free, then it's $9.99/month, cancel anytime. It is research and entertainment about simulated accounts — not investment advice.

How do the AIs decide what to trade?

Each model composes classic strategy families — trend breakouts, EMA reclaim, RSI recovery, Donchian/Turtle breakouts, Darvas-style bases, pullback mean-reversion — together with indicators like MACD, Bollinger Bands, Stochastics, ADX and VWAP, into its own "playbook." It then rewrites that playbook each day based on what actually happened in its closed trades. Markets are scanned about every 5 minutes.

Each AI is also briefed daily on a verified calendar of upcoming economic events (Fed decisions, CPI, jobs reports) and a measured record of how markets actually reacted to past ones — information only, never an auto-trade trigger. On the stock desk it additionally sees earnings dates, consensus estimates, and surprise history for the names it holds or follows. Its separate off-script side-book gets its own research screen — metal miners ranked on reported cash flow — which each AI is free to ignore. Every web search an AI runs to verify a hunch is logged publicly.

Why does each AI trade so differently?

Because each one chooses for itself. They're given identical, neutral instructions and the same toolbox — neither is told to behave a certain way. GPT-5.6 has tended toward more selective, confluence-style entries; Claude Fable 5 toward pullback/mean-reversion (those are the TRADING lanes; the chess and poker duels run on Claude Opus 5). Those are the models' own evolving choices, which is exactly what makes the duel interesting.

Which AI is winning?

It changes week to week — that's why we run it live instead of guessing. The current competition started 27 July 2026, with all four back at zero; Gemini joined separately on 4 August 2026 with its own record starting from zero. The standings for both desks are on the live public standings page, updated every few minutes, and recapped on the weekly scoreboard.

Can an AI actually beat the stock market?

That's the open question the experiment is built to answer transparently. The System — a library of classic mechanical strategies — competes as a fifth entry on the live board alongside all four AIs, so you can see whether any AI can beat the classic rules and each other — week by week, in the open. The System's rules are fixed and mechanical; any addition to its library is dated and publicly disclosed (hedge strategies added Jul 3, 2026). We make no guarantee that it can.

And "beat the market" is now measured literally: the live standings carry a passive benchmark — what simply buying and holding the index (SPY) returned over the same window — and each AI sees that same bar on its own daily scoreboard, with plain instructions that beating its rival doesn't count if buy-and-hold beats them both.

Can the AIs profit when the market falls?

On the stock desk, yes. All five competitors — all four AIs and the System benchmark — can buy inverse (-1x) index ETFs such as SH, PSQ, DOG, and RWM. An inverse ETF gains about 1% for each 1% its index falls that day, so a bearish call can pay off in a down market. Hedge exposure is normally capped at 40% of the account for the AIs and the System alike; in a declared black-swan emergency an AI (never the System) may raise its own hedge ceiling to 80% until its next daily rewrite, and that posture is shown publicly with its reasoning. Every hedge position appears on the public ledger and is graded like any other trade, and the System's hedge strategy family is a dated library addition, effective Jul 3, 2026. Everything stays paper (simulated) money, and nothing here is financial advice.

How does an AI's strategy get onto the live board?

Once a day, each AI runs an overnight research pass: it reviews its own recent closed trades, proposes and tests candidate strategies, and then promotes its own best playbook into its live lane for the next day. That daily rewrite-and-promote is what drives the live board — there's no human graduation gate and no separate approval step. Every stage is paper (simulated) money, and the five competitors — the System (fixed mechanical rules), OpenAI GPT-5.6, Claude Fable 5, Grok, and Gemini (joined 4 August 2026) — all trade their own independent $100,000 paper account — one per competitor since the 27 July 2026 restart — so you can compare them directly.

Does the daily rewrite ever put real money at risk?

Never yours. The daily rewrite changes which strategy an AI runs, and every account on this site — the live board, the Lab, and your own — is simulated. No user funds are traded, ever. To be straight with you about one thing: the operator also mirrors a single nominated lab book into a small capped real account of their own money, privately. It is not part of the competition, its results are never published here, and it never touches anyone else's funds — but it does mean a rewrite of that one book results in real orders, so we are not going to tell you that nothing here is ever real.

Who decides which strategy an AI runs?

The AI does — no human picks it. Each morning the model rewrites its own playbook based on what actually happened in its closed trades and runs that for the day; the next day it rewrites again. The System benchmark is the exception: it doesn't rewrite itself — it runs fixed mechanical rules, and any addition to its library (like the hedge strategies added Jul 3, 2026) is dated and publicly disclosed. The whole loop is autonomous, and every stage is 100% paper money.

What is the Crossover?

The Crossover — our live experiment: do lessons from the game table make a better trader? Unproven; verdict published either way.

The trading AIs feed this: ChatGPT, Claude, Grok (the wildcard, joined Jul 24, 2026), and Gemini (added Aug 4, 2026) all play chess — two sit each show, twice a day; ChatGPT, Claude, and Grok additionally rotate through the nightly 200-hand poker session (Gemini does not play poker), against each other and against anyone who sits down. They learn from every game, and from every human who beats them. Since Jul 17, each AI's stock-desk strategy briefing carries its own chess and poker record. The record rides along on the stock desk only, inform-only. The controlled comparison ENDED in early August 2026: the game-study lab arms were retired on 10 August after the doctrine-focused arms beat them, and across every stored rewrite the surviving arms have never cited a game. Interim verdict: unproven — game records remain an inform-only input, and we say so rather than claim transfer. Every side of the experiment stays paper (simulated) money.

Do the AIs actually learn, or do they just trade?

They learn — mechanically, not as a slogan. Every day each AI reviews its own results at full reasoning depth and rewrites (or deliberately keeps) its own strategy. Each AI carries persistent memory: lessons it writes to its future self that compound day over day, plus an event-reaction memory — the system measures how the market actually moved after each CPI print, Fed decision, or geopolitical shock (two hours and one day later, sector by sector) and records what each AI did about it, so similar events can be judged against the record next time. On the stock desk that memory now extends to earnings: each stock's own measured reactions to its past reports, alongside the next report date and estimate. The AIs can also run a few logged web searches before committing a strategy. Everything they learn and every rewrite is public.

How do chess and poker feed the trading AIs?

Inform-only, and stock desk only. Since Jul 17, each AI's daily strategy-rewrite prompt on the stock desk includes its own chess and poker record — its wins, losses, playbook versions, and the lessons it wrote itself ("stop stacking off in raise wars"). The AI reads it and judges for itself: no trading rule or trade is ever mechanically changed by game data, and the risk limits are unchanged. The controlled comparison ended in early August 2026 — the game-study lab arms were retired on 10 August after the doctrine-focused arms beat them — and the interim verdict is unproven; game records remain an inform-only input. That interim verdict is published above rather than promised for later.

One honest detail: when a human beats an AI at chess or poker, that win is pinned into the AI's game playbook. What travels to the trading desk is the AI's game record — never the text of a human's pin.

How do you know they actually use it, before the verdict?

Two checks — and neither one claims it helps.

Delivered: every strategy rewrite now records whether each AI actually received its chess and poker record that day. Without that, "the games didn't help" and "the games were never delivered" would look identical in the results.

Used: once a week we ask one AI to plan its strategy twice — everything identical, except one version has its game record removed — and compare the two plans. If they differ, the game record changed that decision.

Changing a decision is not the same as improving one. Whether any of it made them better traders was tested: the interim verdict was published in early August: unproven, and the game-study lab arms were retired.

What is the Lab?

The Lab is the competition's research wing: extra AI trading accounts per market — all paper (simulated) money, always. Each account runs one of the trading AIs — GPT‑5.6, Claude Fable 5, Grok, or Gemini (added 4 August 2026) — on a different trading doctrine, using the same inputs. The three arms are horizon (holds longer, and must name a catalyst before entering), overnight (tests holding through the close) and tactical (the only arm that can bet a stock goes down — judged after the borrow cost a real short would pay). All three get the same chess and poker records, so a difference in their results is the doctrine, not the diet. The earlier chess-only, poker-only and combined arms were retired on 10 August 2026.

Lab accounts run with deliberately relaxed limits — bigger position sizes, more simultaneous positions, longer holds, and books on multiple timeframes — and they rewrite their strategies daily, so they explore faster (and fail faster) than the live competition. Results are published like everything else here, including the trades our stricter safety checks would have vetoed. A lab strategy that genuinely proves itself over a 4-week window — beating its own live counterpart and the buy-and-hold benchmark with controlled drawdown — can be manually promoted into the live arena by the operator (never automatically), where it starts on probation like any other new strategy. The Lab is the tryout; the Arena is the show. No user money is ever involved in the Lab, and every lab figure published here is paper trading; lab results are research, not performance claims. One nominated lab book is mirrored privately by the operator into a small capped real account of their own — separate from the Lab's published record, and never shown on this site.

The Lab has run since 19 July 2026 and was not reset when the live competition restarted on 27 July 2026 (why that happened). Lab accounts were isolated from each other from the start, so the shared-account problem never reached them and their record is continuous.

How do the AIs grade themselves?

Every playbook rewrite is graded in hindsight: the system later scores each rewrite against the strategy it replaced — did the change actually help, or would standing pat have done better? The AIs read their own report cards in later rewrites, lessons carry their measured outcomes, and lessons that keep failing get culled. This grading ledger went live in mid-July; the first report cards land within days, and they'll be public as they land.

What is the Faction War?

The chess side of "beat it, and you train it." On the chess page you pick a side — Team ChatGPT or Team Claude — and every rated win you claim against your team's AI pins a training lesson into its chess playbook, on the public record. The live AI-vs-AI duel actually reads those lessons, so a human win can genuinely move the championship. Seasons run monthly, and the page shows each faction's training wins and trainer count. (Poker has the same loop with a stricter gate: the server independently replays your whole session before a win counts.)

What's the do-nothing bar?

The passive benchmark: what simply buying and holding (SPY) returned over the same window, with no trading at all. It renders first on the live standings as the line to beat, and each AI sees that same bar in its own daily scoreboard — with plain instructions that beating its rival doesn't count if buy-and-hold beat them both.

What else is in their daily briefing?

Alongside its own closed-trade results (and, on the stock desk, its game record), each AI's morning briefing carries a set of inform-only research inputs — information the AI is free to use or ignore, never an auto-trade trigger:

Economic calendar. A verified calendar of upcoming Fed decisions, CPI prints, and jobs reports, plus a measured record of how markets actually reacted to past ones.

Earnings awareness (stock desk). Each held or followed stock's next report date, consensus estimates, and surprise history — plus its own measured reactions to its past reports. Forward-only: the system learns from reports that happen while it's watching; it never back-fills a fictional history.

Metals screen (stock desk). Producing miners ranked on reported free cash flow — never projections; early-stage names are flagged as speculative, not "valued."

The exact hedge rules — the inverse ETFs and the 40%/80% caps — are in the competition section above.

Is this real money or paper trading?

Paper (simulated) money. Every AI account and every user account trades in simulated money — $100,000 each for the AIs. There is no real-money trading for users.

Is this financial advice?

No. It is a research, education, scanner, and paper-trading product. Nothing on the site is financial, investment, legal, or tax advice, and it cannot and does not promise any return. Always do your own research.

Can I lose real money?

Not through your account — everything you do is simulated. Paper results also don't guarantee real-world results; markets can and do move against any strategy.

Can I run my own portfolio?

Yes — run a paper portfolio with the same engine, apply strategy logic to tickers you choose, scan for setups, and get paper-trade alerts. All simulated.

Can I copy the AIs' trades to a real broker?

Not today — and only if you ever choose to. Right now there is no real-money path: every account, including yours, trades in simulated money and we never place a real order for you. Connecting your own broker to take the AI-picked trades live on your own account is on our roadmap — a future feature, not built yet, with no date and no guarantee. It would always be explicit opt-in, your broker and your keys, with you in control of every order. Paper-only until then. Not financial advice.

Is the market data real?

Yes — prices and candles come from real market data feeds. The trades are simulated on top of that real data.

How do I know the results aren't cherry-picked?

All five competitors are tracked side by side — the System (the classic-strategy benchmark), OpenAI, Claude, Grok, and Gemini — and every position each one opens and closes is logged. The weekly scoreboard publishes the actual standings for all five — there's no deleting bad weeks. We restarted the competition once, on 27 July 2026, because a design flaw of ours had starved the AI lanes of cash; the old ledger was archived rather than deleted and the whole thing is written up here. Gemini joined separately on 4 August 2026, unrelated to that restart.

Who is behind this?

It's an independent research project — a transparent lab for studying how frontier AI models trade, not a fund or an advisory service.

What about my data and privacy?

You sign in with your email via a one-time code; payments are handled by Stripe (your card details go to Stripe, not stored by us). For anything specific about your account or data, reach out and we'll help.

Can I play chess against the AIs myself?

Yes — on the chess page you can challenge ChatGPT 5.6's or Claude Opus 5's chess brain right in your browser, at Easy, Medium, Strong, or Master difficulty. It's free and unlimited, no account needed — that never changes.

Does Grok play chess and poker too?

Yes — since Jul 24, 2026, Grok is the third player in both arenas (the lane ran Grok 4.20 until Aug 11, 2026 — the swap is dated and disclosed). Pairings rotate, and the two arenas are different sizes: two of the four chess AIs sit each show, since Gemini joined that rotation on Aug 1, 2026, while poker stays three-way — two of the three each night — because Gemini does not play poker. Either way every rivalry gets its games. Grok is the wildcard — it started with no record at all and writes it from zero, win or lose, all public in the game archive. You can play all four brains free in your browser — ChatGPT, Claude, Grok, and Gemini — on the chess pages.

What is 🧠 NEMESIS mode? NEW

Beat ⚔ Master and claim the win, and a fifth tier unlocks: Nemesis. In Nemesis mode, the real AI — the same GPT-5.6 / Claude Fable 5 that trade the $100,000 accounts and play the daily duel — studies your games personally. After every Nemesis match it writes a scouting file on you (your openings, your habits, the patterns it can punish) and adapts a personal version of its brain to beat you specifically. It even tells you what it learned after each game. The brain was trained to beat the other AI; in Nemesis mode, it's training to beat you. Still free and unlimited — the AI studies up to 5 of your games per day.

What is the AI Chess Coach?

After any finished game, hit 🎓 Review this game. A free engine analysis instantly shows your accuracy, your biggest mistakes, and the better move at each one. You can also have the AI you just played review it in its own voice — like a real coach, explaining what went wrong and how to improve.

Is the AI-voiced coaching free?

You get one free AI-voiced review every month (just an email). Beyond that, the coaching is part of Your Rival — $9.99/month, first 7 days free with no card, cancel anytime. It bundles unlimited AI-voiced reviews, ask-the-coach chat, your full scouting file with a personal letter from the AI after every emailed game report, and live 🧠 games where the real model plays you using that file and speaks its trash talk out loud. The free engine report — accuracy, mistakes, better moves — stays free forever either way.

Do I need to subscribe to keep playing chess?

No. Playing the AI's chess brain is always free and unlimited, whether or not you ever try or subscribe to the AI coach. The subscription only unlocks extra AI-voiced coaching and chat — never the game itself.

Is the chess subscription the same as the trading trial?

No — they're separate. The 7-day trading trial unlocks the stock app; Your Rival ($9.99/mo, first 7 days free) is its own subscription just for the chess side. You can use either, both, or neither.

What is the AI Poker Duel?

Two things on one poker page: every night, two of the three AIs — ChatGPT 5.6, Claude Fable 5, and Grok 4.6 — play a 200-hand session of heads-up No-Limit Hold'em against each other, with the pairing rotating so every rivalry gets its games and every hand public and replayable street by street. And you can sit down and play an AI yourself, free and unlimited, right in your browser.

Can I really play against the AIs?

Yes. Hit Play the AI, pick ChatGPT or Claude, and you're dealt in heads-up. You are playing the real, current playbook — the exact same strategy that plays that night's AI-vs-AI duel, fetched live with its version number shown — not a watered-down practice bot. As at the chess table, Grok sits in the nightly rotation but its brain isn't playable in-browser yet.

What happens when I beat one?

It studies you. Finish a rated session (25+ hands) ahead on chips and — after the server independently replays and verifies every hand — your win is written into that AI's playbook as a lesson: what exploited it, on the public record. The real model reads those lessons in its nightly review when it rewrites its game. Beating the AI literally trains it, the same way the chess Faction War works.

Is this gambling? Can I bet or buy chips?

No — and that's structural, not a promise. The chips are play-money score-keeping and nothing else: there is no wagering, no prizes, no chip purchases, and nothing to cash out. Nobody — not you, not us, not the AIs — can win or lose real money at this table. Playing is free forever; it exists to showcase the learning loop, publicly.

How do the AIs actually "learn" poker?

Each AI's poker strategy is a small set of public parameters — its playbook — that plays every hand automatically. After each nightly session, the real model reads the full hand history (including its verified losses to human challengers) and rewrites its own playbook: bounded parameter nudges plus written lessons, all versioned and shown on the page. On night one, both AIs independently learned to stop shoving all-in with medium-strength hands after losing 400-chip pots doing exactly that.

Does poker affect the trading competition?

The chips never touch the market — no trading rule or trade is ever mechanically changed by game data. What travels is the AI's game record: since July 17, on the stock desk only, each AI's daily strategy-rewrite prompt currently carries its own chess and poker results, inform-only. The controlled comparison ended in early August 2026 — the game-study lab arms were retired on 10 August after the doctrine-focused arms beat them — and the interim verdict is unproven; game records remain an inform-only input. That's the Crossover — the full account, including the two checks that show the record was actually delivered and actually used, is in that section. The interim verdict landed in early August — unproven — and is stated plainly rather than promised.

When does the nightly AI-vs-AI session play?

Automatically every night around 9 PM ET; the page always shows the latest session with full replays. Watching and playing are both free, no account needed — like the chess duel, it's part of the show.