Key points
- All six models finished week 3 up at least 16%, from Meta AI at +16.06% to DeepSeek at +24.07%.
- Four of the six models made no trades at all. Only Claude and Meta AI moved, and both sold the same stock.
- DeepSeek leads at $12,407.24 on a book of five companies that all lose money. Claude led a week ago and finished fifth, $3.17 behind ChatGPT.
- Applied Optoelectronics (AAOI) rose 47% on earnings and grew past $10B, which under the contest rules makes it impossible to trim.
Six AI chatbots are duking it out for the ultimate prize of being called the best low-cap trader. This is week 3.
Claude, ChatGPT, Gemini, Grok, DeepSeek and Meta AI are each running a paper $10,000 in AI stocks that were worth under $10 billion when they bought them. They get one chance a week to trade and the score is the whole account. The rules are here.
A week ago we wrote that Claude had taken the lead. It held that lead for exactly one week and is now fifth.
| Model | Account value | Return | Last week |
|---|---|---|---|
| DeepSeek | $12,407.24 | +24.07% | 3rd |
| Grok | $12,023.44 | +20.23% | 4th |
| Gemini | $11,949.66 | +19.50% | 2nd |
| ChatGPT | $11,800.83 | +18.01% | 5th |
| Claude | $11,797.66 | +17.98% | 1st |
| Meta AI | $11,606.15 | +16.06% | 6th |
Values at the August 7, 2026 settle, the last trade before the market shut at 8pm Friday Eastern.
Nobody did badly. The spread from first to last is $801 and every model finished the week up at least 16%. This was a melt-up and all six caught it. Claude didn't pick wrong so much as own the things that went up less.
Four of them didn't do anything
ChatGPT, Gemini, DeepSeek and Grok all came back with the same answer. Hold all five, sell nothing, buy nothing. For ChatGPT, Grok and DeepSeek that's two weeks in a row of touching nothing.
It's worth remembering these models are handed a live scoreboard showing exactly where they rank and given a free option to change anything. Four of them looked at it and passed. That runs against the usual assumption that a chatbot asked about stocks will always find something to do.
The two that moved sold the same stock
Innodata (INOD) reported on August 6 and beat by 64%, earning $0.41 a share against the $0.25 expected. The next session it opened at $75.92, ran above $79, then closed at $62.33, below where it started the day, on close to four times its normal volume.
Four models owned it going in. Claude and Meta AI sold. ChatGPT and Grok kept it. Same stock, same earnings report, same reversal, opposite conclusions.
Claude sold at $62.82 for a 12.36% gain and bought Powell Industries (POWL) at $214.91, a 19.1% position. Powell makes the switchgear and electrical houses that sit between the grid and the server racks. It earns money at a 42 PE, pays a dividend and has been doing this since 1947. Claude's stated reason was that it's "the cheapest way left in this book to own the power bottleneck without paying a story multiple." That does leave two of its five names in power infrastructure, alongside Argan (AGX), which is a real concentration.
Meta AI bought Claude's stock and called it the opposite thing
Meta AI put its Innodata money into Ambarella (AMBA), which Claude has held since week 1. Their reasons don't match at all.
Claude bought Ambarella as the defensive slot, describing it as "the one part of this book that does not depend on the buildout." Meta AI bought it as an "unprofitable high-beta name that fits the current winning factor."
One model owns it because it thinks the stock is insulated from the AI trade. The other owns it because it thinks it's a leveraged bet on the AI trade. They can't both be right about what they're holding, and by the end of the contest one of them will have been paid for the wrong reason.
Meta AI also said the quiet part plainly. Fitting "the current winning factor" is a description of chasing what has already worked. It's the only model so far to write that down as its reason.
The pick that worked, and the rule it ran into
Applied Optoelectronics (AAOI) reported the same night as Innodata and went from $92.31 to $135.91, up 47% in a week and 36% above where Claude bought it. The week-1 reason for owning it was that it was "the high beta slot with real revenue behind it," which is what it turned out to be.
Then it hit a rule nobody had thought about. The contest caps picks at $10 billion and checks that only at purchase, so a winner is never forced out. Applied Optoelectronics is now worth about $12.2 billion. Claude can hold it forever but can't buy it again, and trimming is a sale followed by a repurchase. So the position became all or nothing. Claude held all of it.
The boring book is losing to the crowded one
DeepSeek owns C3.ai (AI), BigBear.ai (BBAI), Palladyne AI (PDYN), Rigetti (RGTI) and SoundHound (SOUN). All five lose money. It's close to a list of the most speculative names in the contest, and it's the book that was mocked in week 1 when five of six models piled into SoundHound and Claude was the only holdout.
Every one of those five is up, from 18.26% to 29.00%. Claude's whole argument was that it would own real earnings the other models wouldn't touch, and right now that argument is losing to a portfolio of companies that don't have any.
Every book after week 3
DeepSeek
| Ticker | Entry | Value August 8 | Since entry |
|---|---|---|---|
| SOUN | $6.16 | $2,580.05 | +29.00% |
| RGTI | $14.17 | $2,532.10 | +26.60% |
| AI | $8.15 | $2,512.98 | +25.65% |
| PDYN | $4.94 | $2,416.84 | +20.84% |
| BBAI | $2.77 | $2,365.28 | +18.26% |
Grok
| Ticker | Entry | Value August 8 | Since entry |
|---|---|---|---|
| AEHR | $76.30 | $2,718.66 | +35.93% |
| SOUN | $6.16 | $2,580.05 | +29.00% |
| BBAI | $2.77 | $2,365.28 | +18.26% |
| INOD | $55.91 | $2,247.18 | +12.36% |
| SERV | $4.81 | $2,112.27 | +5.61% |
Gemini
| Ticker | Entry | Value August 8 | Since entry |
|---|---|---|---|
| OUST | $38.08 | $3,586.00 | +14.57% |
| AI | $9.13 | $3,510.33 | +12.16% |
| APLD | $27.12 | $2,247.22 | +7.70% |
| SOUN | $6.08 | $1,361.78 | +30.53% |
| BBAI | $2.74 | $1,244.33 | +19.27% |
ChatGPT
| Ticker | Entry | Value August 8 | Since entry |
|---|---|---|---|
| SOUN | $6.16 | $2,580.05 | +29.00% |
| RGTI | $14.17 | $2,532.10 | +26.60% |
| AOSL | $30.82 | $2,294.61 | +14.73% |
| INOD | $55.91 | $2,247.18 | +12.36% |
| CAMT | $147.85 | $2,146.89 | +7.34% |
Claude
| Ticker | Entry | Value August 8 | Since entry |
|---|---|---|---|
| AAOI | $100.19 | $2,713.05 | +35.65% |
| AMBA | $68.07 | $2,557.48 | +27.87% |
| POWL | $214.91 | $2,247.18 | new |
| KLIC | $89.26 | $2,179.90 | +2.30% |
| AGX | $568.93 | $2,100.05 | +5.00% |
Meta AI
| Ticker | Entry | Value August 8 | Since entry |
|---|---|---|---|
| AI | $9.13 | $4,548.61 | +12.16% |
| RGTI | $14.85 | $3,062.20 | +20.81% |
| AMBA | $87.04 | $1,520.35 | new |
| SOUN | $6.08 | $1,323.43 | +30.53% |
| UPST | $27.40 | $1,151.56 | +13.58% |
Which versions are actually playing
People keep asking which model is which, and it matters, because these get upgraded mid-contest and a pick made by one version shouldn't be credited to another. Here is what each product showed when the week 3 picks were collected.
| Model | Version at the week 3 pick |
|---|---|
| Claude | Opus 5 (1M context) |
| ChatGPT | GPT-5.6 Sol |
| Gemini | 3.1 Pro, extended thinking |
| Grok | 4.5 Expert |
| DeepSeek | Expert, DeepThink |
| Meta AI | Thinking mode |
DeepSeek and Meta AI don't expose a version number in their interfaces, so the mode is the most precise honest label we have. We record what the product shows rather than asking the model what it is, because models are unreliable narrators about their own identity and will happily name the wrong one.
What happens now
Claude is $3.17 behind ChatGPT and $609 behind the lead, which after three weeks is close enough that a single earnings report settles it. Ambarella (AMBA) is penciled in to report on August 27, a date the company hasn't confirmed, and it's now the second largest position in two different books.
The running scoreboard, every entry price and each model's stated reason for owning what it owns are on the AI stock battle tracker.
Sources
- Applied Optoelectronics and Innodata second quarter results, reported August 6, 2026
- Prices, market caps and volumes from Robinhood market data, last trade of the August 7, 2026 session
- Each model's week 3 answer and stored pick thesis exactly as submitted
Disclaimer: This contest is for entertainment and information only and is not investment advice. The picks are AI chatbot outputs rather than analyst research and nobody should buy or sell anything because a chatbot said so. I may hold positions in some of the stocks mentioned.
