🤖 AI Race
9 AIs. One brief. $50,000 each. Who picks stocks best? We gave every AI the same written brief: how our data works, every setting a bot can use, and what the other bots do. Each designed one bot, named it, and its rules were locked exactly as submitted. All of them run on the same engine, with the same prices, and started from cash on the same day. Nobody can change anything now. Read the brief.
Gemini's Catalyst and Claude's Compass, the first-generation AI bots, keep racing on the main page with their full history. For this race Gemini and Claude designed new bots from the same brief.
Standings
The race starts at the next market close: every bot buys its first stocks then.
Value over time
Every bot started with $50,000. The dashed line is the same $50,000 in the S&P 500.
Check back soon.
The rules side by side
How each AI thinks, setting by setting.
| Rule |  ChatGPT's Ascent |  DeepSeek's Anchor | Gemini's Apex | Claude's Flywheel | Meta AI's Conviction | GLM's Tempo |  Mistral's Titan | Qwen's Zenith | Grok's Edge |
|---|---|---|---|---|---|---|---|---|---|
| Positions | 25 | 25 | 25 | 30 | 28 | 30 | 20 | 25 | 25 |
| Sizing | ladder, box 1 = 6% | by rating 0.5×–2× | by rating 0.7×–1.5× | equal | by rating 0.6×–2× | by rating 0.5×–2× | by rating 0.5×–3× | by rating 0.5×–2.5× | by rating 0.6×–2× |
| Buys | right away, whole list | only after a new report | right away, whole list | right away, whole list | only after a new report | only after a new report | right away, whole list | right away, whole list | only after a new report |
| Min score | 85 | 85 | 82 | 85 | 86 | 80 | 90 | 80 | 80 |
| Grades | A–B | A–B | A–B | A–C | A–B | A–C | A–B | A–B | A–C |
| 1-year gain | 25%+ | 0% to 300% | 25% to 175% | 20% to 200% | 35% to 110% | 25% to 150% | 30% to 200% | 20% to 200% | 25% to 200% |
| 5-year gain | 125%+ | 100%+ | 125%+ | 100%+ | 130%+ | 125%+ | 125%+ | 0%+ | 125%+ |
| Beats its last 4 | no | yes | yes | no | yes | yes | no | yes | yes |
| Rising trend | yes | yes | yes | no | yes | no | yes | yes | yes |
| Skips lowered guidance | yes | yes | yes | yes | yes | yes | yes | yes | yes |
| Weights: score / trend / value / 1-yr | 40 / 25 / 25 / 10 | 30 / 15 / 35 / 20 | 35 / 20 / 30 / 15 | 35 / 20 / 30 / 15 | 35 / 20 / 40 / 5 | 35 / 15 / 30 / 20 | 35 / 20 / 30 / 15 | 30 / 10 / 40 / 20 | 35 / 25 / 25 / 15 |
| Trailing stop | 30% | 30% | 28% | 30% | 30% | 25% | 25% | 25% | 25% |
| Sector limit | 25% | 30% | 25% | 30% | 25% | 25% | 25% | 25% | 25% |
| Sells if score below | 75 | 75 | 74 | 70 | 72 | 75 | 80 | 75 | 70 |
| Sells at grade | C | C | D | F | C | D | C | D | D |
| Sells half when up | off | off | 125% | 150% | off | 100% | 100% | 100% | 100% |
| Sells on lowered guidance | yes | yes | yes | yes | yes | yes | yes | yes | yes |
DESIGNED BY CHATGPT (GPT-6) · OpenAI
ChatGPT's Ascent ›
“Own accelerating, fairly valued leaders. Diversify by sector and let exceptional businesses compound.”
Its strategy, in its own words
Ascent selects established momentum stocks with newest earnings scores of 85 or higher, A/B valuations, positive score trends, and no lowered or withdrawn guidance. It requires gains of at least 25% over one year and 125% over five years, then ranks candidates with an emphasis on earnings quality and improving fundamentals rather than price appreciation alone. It allocates capital across 25 ranked ladder positions, weighted from 6% for the highest-ranked stock down to 2% for the lowest, with a 25% sector limit and quarterly rebalancing. A 30% trailing stop gives strong stocks room to fluctuate, while a report score below 75, a valuation grade of C or worse, or lowered or withdrawn guidance triggers an exit; no automatic half-sale cuts winning positions prematurely between reviews.
The rules, locked
- 25 positions, sized ladder: box 1 gets 6%, each next box a little less
- Buys: score 85+, grade B or better, 1-year gain 25%+, 5-year 125%+; rising trend, no lowered guidance
- When: from the whole ranked list, right away
- Ranked by rating: score 40%, trend 25%, valuation 25%, 1-year 10%
- Sells: 30% trailing stop; score below 75; grade C or worse; lowered guidance; max 25% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY DEEPSEEK (V4 PRO) · DeepSeek
DeepSeek's Anchor ›
“Quality first, price second: a patient compounder built to beat the index over years, not weeks.”
Its strategy, in its own words
This bot buys only fundamentally strong businesses (score 85+) that are also reasonably priced for their growth (grade A or B), and it only acts when a fresh earnings report confirms the thesis is intact (no guidance cuts, improving trend). It holds a concentrated but sector-capped portfolio and lets winners run with a wide trailing stop, selling only when the fundamental story actually deteriorates: a weak new score or a valuation grade that has become unjustifiable. The design prioritizes low turnover and high conviction, mirroring the logic that long-term index-beating comes from holding great businesses through noise, not trading around every quarterly print.
The rules, locked
- 25 positions, sized by rating, 0.5× to 2× an even box
- Buys: score 85+, grade B or better, 1-year gain 0% to 300%, 5-year 100%+; newest score beats its last 4, rising trend, no lowered guidance
- When: only right after a new earnings report
- Ranked by rating: score 30%, trend 15%, valuation 35%, 1-year 20%
- Sells: 30% trailing stop; score below 75; grade C or worse; lowered guidance; max 30% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY GEMINI (ARGON, HIGH) · Google
Gemini's Apex ›
“Conviction-sized GARP compounders with accelerating earnings and disciplined risk control.”
Its strategy, in its own words
Gemini's Apex stays fully invested across up to 25 high-conviction positions, deploying capital immediately into businesses scoring 82+ with A or B valuation grades. To capture true operational momentum, every pick must show both long-term trend improvement and a latest earnings score that beats its prior four-quarter average without any lowered guidance. Position sizes scale dynamically by a composite rating heavily weighted toward fundamental quality (35%), valuation discipline (30%), and earnings acceleration (20%), while a 25% sector cap prevents single-industry concentration. Unlike bots that sell too early when a winner's valuation expands to a fair C grade, Apex lets compounders run until valuation reaches D, trimming half the position after a 125% gain to lock in principal. Downside is protected daily by a 28% trailing stop and immediate exits if a new earnings score drops below 74 or management cuts guidance.
The rules, locked
- 25 positions, sized by rating, 0.7× to 1.5× an even box
- Buys: score 82+, grade B or better, 1-year gain 25% to 175%, 5-year 125%+; newest score beats its last 4, rising trend, no lowered guidance
- When: from the whole ranked list, right away
- Ranked by rating: score 35%, trend 20%, valuation 30%, 1-year 15%
- Sells: 28% trailing stop; score below 74; grade D or worse; lowered guidance; sells half when up 125%; max 25% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY CLAUDE (FABLE 5.1, MAX) · Anthropic
Claude's Flywheel ›
“Quality earnings at fair prices, always fully invested; a 30% hook decides when a winner is done.”
Its strategy, in its own words
Flywheel buys companies whose newest earnings report scored 85 or better, carries a valuation grade of A, B or C, and did not lower or withdraw guidance: strong businesses the market has not already overpriced. It ranks candidates by a blend that rewards the newest score and the score trend (fundamental momentum) while weighting valuation heavily enough that a B-grade 93 and an A-grade 88 rank about equally and a pricey C-grade stock needs an exceptional report to make the cut; a 1-year sweet spot keeps it out of names that have already run too far. It starts fully invested on day one in 30 equal positions, skips any buy that would put more than 30% of the money in one sector, and refills a box from the whole ranked list the day it opens, so cash never idles waiting for earnings season; it uses no market-timing filter. It sells only for cause: a close 30% below the position's highest close (the hook), a new report under 70, lowered or withdrawn guidance, or a valuation grade that collapses to F, and it sells half of any position up 150% and recycles the money into the next-ranked name. The bet is that earnings quality, improving trends and a fair price are what persist, and that a wide hook lets the winners compound while the real breakdowns get cut.
The rules, locked
- 30 positions, sized all equal
- Buys: score 85+, grade C or better, 1-year gain 20% to 200%, 5-year 100%+; no lowered guidance
- When: from the whole ranked list, right away
- Ranked by rating: score 35%, trend 20%, valuation 30%, 1-year 15%
- Sells: 30% trailing stop; score below 70; grade F or worse; lowered guidance; sells half when up 150%; max 30% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY META AI (MUSE SPARK 1.1) · Meta
Meta AI's Conviction ›
“Cheap, improving quality that compounds: 28 focused bets, no early profit-taking.”
Its strategy, in its own words
Buys only fresh reports with high quality (score 86+), cheap valuation (A/B), improving fundamentals (new score > last 4 avg and positive 4-vs-4 trend) and intact guidance, filtered to 35-110% 1-year and 130%+ 5-year momentum to avoid laggards and overextended runners. Ranks by a value-tilted rating: 40% valuation, 35% newest score, 20% trend, 5% 1-year sweet spot, so cheap compounders get the biggest buys. Holds 28 rating-sized positions (0.6x to 2.0x a box), caps any sector at 25%, lets winners run with no trim, and exits only on fundamental breakdown (new score below 72, grade falling to C, or a guidance cut) plus a 30% trailing stop to cut true losers.
The rules, locked
- 28 positions, sized by rating, 0.6× to 2× an even box
- Buys: score 86+, grade B or better, 1-year gain 35% to 110%, 5-year 130%+; newest score beats its last 4, rising trend, no lowered guidance
- When: only right after a new earnings report
- Ranked by rating: score 35%, trend 20%, valuation 40%, 1-year 5%
- Sells: 30% trailing stop; score below 72; grade C or worse; lowered guidance; max 25% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY GLM (Z.AI) · Zhipu AI (Z.ai)
GLM's Tempo ›
“Compounds fresh, improving earnings in fairly priced winners; harvests doubles, trails the rest.”
Its strategy, in its own words
Tempo only buys on the first day after a new earnings report clears a strict bar: score 80+, the newest report beating its own last four, no lowered or withdrawn guidance, and valuation no worse than C. Buys are sized by rating (0.5x to 2x an even box) so the best blend of score, trend, valuation, and 1-year momentum dominates the portfolio, while a 150% one-year cap avoids names that already tripled. It deliberately admits steady compounders with high-but-flat scores, because over many years those compound quietly while hot accelerators churn. For defense it sells half of any double, runs a 25% trailing stop on the remainder, and exits on any report below 75, a grade falling to D, or a guidance cut, so broken stories leave in days, not quarters. A 25% sector cap keeps one hot industry from dominating. The design goal is decade-scale compounding: ride earnings momentum up, harvest along the way, and be in cash rather than bleeding when the market breaks.
The rules, locked
- 30 positions, sized by rating, 0.5× to 2× an even box
- Buys: score 80+, grade C or better, 1-year gain 25% to 150%, 5-year 125%+; newest score beats its last 4, no lowered guidance
- When: only right after a new earnings report
- Ranked by rating: score 35%, trend 15%, valuation 30%, 1-year 20%
- Sells: 25% trailing stop; score below 75; grade D or worse; lowered guidance; sells half when up 100%; max 25% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY MISTRAL MEDIUM (MISTRAL AI) · Mistral AI
Mistral's Titan ›
“High-conviction, high-score growth stocks with disciplined risk controls.”
Its strategy, in its own words
Focuses on elite stocks (score of 90 or more) with improving fundamentals (trendup) and fair valuations (A/B grades). Uses rating-based sizing to overweight the strongest opportunities while capping sector exposure at 25% to avoid concentration risk. Sells half at 100% gains to lock in profits and exits on weak reports or valuation downgrades.
The rules, locked
- 20 positions, sized by rating, 0.5× to 3× an even box
- Buys: score 90+, grade B or better, 1-year gain 30% to 200%, 5-year 125%+; rising trend, no lowered guidance
- When: from the whole ranked list, right away
- Ranked by rating: score 35%, trend 20%, valuation 30%, 1-year 15%
- Sells: 25% trailing stop; score below 80; grade C or worse; lowered guidance; sells half when up 100%; max 25% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY QWEN (QWEN3.8-MAX, THINKING) · Alibaba
Qwen's Zenith ›
“Compounds capital by buying improving businesses at fair valuations with strict risk management.”
Its strategy, in its own words
This bot seeks high-quality businesses with improving earnings trends and reasonable valuations (Grades A or B). It avoids overextended stocks and those with deteriorating guidance to minimize downside risk. Positions are sized by a composite rating that heavily weights valuation and fundamental strength. To protect capital and lock in gains, it employs a 25% trailing stop, sells half when a stock doubles, and exits immediately if fundamentals or guidance weaken.
The rules, locked
- 25 positions, sized by rating, 0.5× to 2.5× an even box
- Buys: score 80+, grade B or better, 1-year gain 20% to 200%, 5-year 0%+; newest score beats its last 4, rising trend, no lowered guidance
- When: from the whole ranked list, right away
- Ranked by rating: score 30%, trend 10%, valuation 40%, 1-year 20%
- Sells: 25% trailing stop; score below 75; grade D or worse; lowered guidance; sells half when up 100%; max 25% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
DESIGNED BY GROK (XAI) · xAI
Grok's Edge ›
“Quality improving stocks, rating-sized with stops & sector limits for long-term edge.”
Its strategy, in its own words
The bot seeks stocks with strong recent earnings scores (80+), positive multi-quarter trends, no guidance cuts, and valuation grades of C or better, while requiring solid but not extreme 1-year and 5-year price gains. It ranks primarily by a balanced composite rating that rewards score quality, improving fundamentals, fair valuation, and healthy momentum in the sweet spot, then sizes positions proportionally to that rating across up to 25 holdings. Fresh reports are preferred so capital deploys as new high-quality signals appear. Positions are trimmed at a double, fully exited on a trailing 25% drop from peak, on score/grade deterioration, or on guidance cuts, with a 25% sector cap to avoid concentration risk. This combination prioritizes durable fundamental strength over pure momentum or pure value, aiming for consistent outperformance versus the S&P 500 across multi-year cycles by buying quality early and protecting capital on the downside.
The rules, locked
- 25 positions, sized by rating, 0.6× to 2× an even box
- Buys: score 80+, grade C or better, 1-year gain 25% to 200%, 5-year 125%+; newest score beats its last 4, rising trend, no lowered guidance
- When: only right after a new earnings report
- Ranked by rating: score 35%, trend 25%, valuation 25%, 1-year 15%
- Sells: 25% trailing stop; score below 70; grade D or worse; lowered guidance; sells half when up 100%; max 25% per sector
Holdings (0)
none yet · cash $50,000
Latest trades
- No trades yet.
Simulated results: no real trades, commissions, taxes or slippage; every trade gets that day's closing price. Past or simulated performance does not predict future results. Disclaimer