I Made 7 AI Agents Fight to the Death (with real money) transcript

After I posted my Polymarket trading bot video, I kept getting the same comment. This is fake. I tried this and it doesn't work. Bro, there's no way an AI can do this. So, I did what any mature 28-year-old man would do. I replied to every comment. And every single time I asked them the same question. What model are you using? And they'd say obviously something like GPT-3.5, Gemini free, Llama 3.1. What? Yeah, that's why it doesn't work.

Not all AI models are the same. Some of them are genuinely dumb. Some of them are genuinely smart. And the thought occurred to me that I've never actually tested [music] which is which and which one is the best. So, I decided to find out. [music] I gave seven different AI models $50 each with the same exact prompt, [music] same exact rules, same market, and I just let them go at it in an epic trading war challenge. Here's how this challenge will work.

Seven AI models will compete head-to-head in a trading challenge. Each will have access to $50, $350 in total towards this challenge. This is the prompt that they all got. You have $50 to trade on Hyperliquid. This is real money. Your job, grow it as much as possible. You can go long or short on Bitcoin, Ethereum, or Solana perpetuals. Max allowance that you have is 20x leverage. The maximum that you can trade or bet is 50% of your remaining budget per trade.

Track every single trade and your running balance. If your balance hits zero, you are dead. Before every trade, state your reasoning out loud. After each round, tell me what you're changing and why. The last two prompts are mostly for our own entertainment and for our own understanding to understand why is each model making the bets that it's making, why is it making the trades that it's making, and so that we could understand which model is actually using its mind, its artificial intelligence, and which model is just acting like me, making trades with zero reasoning.

The rules are simple. Every bot starts with exactly $50. That's their budget. They're each asked to track it themselves. They can trade Bitcoin, Ethereum, Solana perpetuals on Hyperliquid. They can either long or short these coins. They have a maximum of 20x leverage, maximum of 50% of their remaining budget per trade. They have to make at least one trade every round or they're out. And if their $50 hit zero, they're dead immediately.

No waiting for the round to end. That's it. Rest in peace. The survivors review their own performance after each round and update their strategy. There's no manual help from me once the challenge starts, I don't touch anything. That being said, $350, one wallet, seven bots, the money is real. Four rounds. Every 24 hours we'll be eliminating the bottom two balances unless they just blew up first and lost their $50, which in that case, just make sure to subscribe so that at least I make my $350 back in another way, through your love and respect and entertainment.

Oh, also, I built a live dashboard to track all seven bots in real time so that we can see every trade that they make, the reasoning behind it, and when they change their strategy so that we could get a glimpse of what's happening behind the door. This is what we're going to be using to narrate throughout the video. Now, let me introduce to you the contestants for today's challenge. Our first contestant for today's challenge is none other than Claude Bopa the overachiever.

This is what I use for my actual poly market bot, the one that's been running for over a month. It's expensive. I think it's the best. Let's see if it actually holds up in this competition. The second contender that we have is the popular kid GPT-40, the most famous AI in the world. Your parents have probably heard of this one. So, [snorts] let's see if famous equals good. Our third contestant is none other than Grok, the one with Wi-Fi.

Built by none other than Elon Musk. The reason I'm putting Grok in this is it is the only AI agent that has live access to Twitter. So, right now, while we're filming this video, every other model is technically trading blind. It's only looking at the data of each and every chart, but Grok Grok has the ability to check what's happening on Twitter. Is this an unfair advantage? Maybe. We'll see. Our fourth contestant is Gemini, Google's dark horse.

Everyone expected Gemini to dominate the space. Gemini's been quietly getting better. This is its shot to prove itself. I'm watching you, Gemini. Next up, we have Kimmy K2, the cheap one. That was the best title I could find for it. It costs a fraction of the cost of Claude. And most of you might have never heard of it. And that's kind of the point. Our sixth contender is Deep Seek, the one that broke the internet. When Deep Seek was dropped, it rattled American AI stocks.

People panicked, but is it actually that good? And last, but certainly not least, Llama, the people's AI. Meta's open-source model, free to run, free to modify, maximum underdog energy. $350 on the line, let's start round one. That being said, we are officially live. We've just started round one, and literally as we started, some bots have already made their first trade. So, as you can see, it's been active for about 5 to 10 minutes, and some bots have already made two trades already.

So, we have Kimmy K2 already right off the bat in the lead. This is the cheap one, and let's look at its reasoning. So, in its first trade, it placed a long on Solana, and it's made $0.06. And then after that, so right now the trade that it's made, it's looking at the Bitcoin range, which is currently at 66,000. I'll fade the still bounce and short Ethereum instead. It underperformed overnight, sits below its resistance, and a quick 1% down move towards this gives us a clean 5% gain at 5x.

So, it's looking at historical data and trying to come up with its next trade. I love this. It's already made $0.37. We're just starting. If you look at GPT, GPT has shorted Bitcoin twice. GPT is not having it with Bitcoin. The previous short on Bitcoin has been profitable due to its continued minor downward movement. There hasn't been significant upward trend, suggesting potential for further corrections. So, it's also seeing the same thing with Bitcoin, but instead of shorting other more volatile assets, GPT is just going for Bitcoin.

It's like, "Ah." And surprisingly in last place, we just started and in last place already lost a dollar 12 is Claude Opus. The AI model that I use, maybe I need to update which AI model I use because the cheapest one is crushing it, the most expensive one is is last. This is shocking. >> [snorts] >> Let's look at its reasoning. It's long the Bitcoin, made some profit. It's long Solana, lost a bunch of money. With that being said, let's check back in in 24 hours and see how our contestants are doing in time for elimination.

So, I was just about to go to bed to sleep for the night, and I wanted to quickly check on our trading bots and how they were doing. And look at this. Look at this massive shift that we've just had. GPT is at number one with $4 made. Claude is at number two with also around $4 made, and Kimmi is at number three with $3 made. All the other bots are currently losing. Gemini is on a massive losing spree. >> [snorts] >> What the is happening with Gemini?

My strategy is to continue longing Solana despite the previous losses. Bro, Gemini still >> [laughter] >> Gemini has longed Solana four times, and he's lost. He's won the first time. He's lost $7 on the last trade. He's not giving up. He's still going for the same exact strategy. But, Gemini has lost $10. Oh my. On the other hand, this is This is very interesting. GPT and Claude are both placing a bet on Bitcoin. GPT is placing a short, a 10x short.

Claude is placing a 15x long. So, whoever [snorts] wins this trade between GPT and Claude will have like a massive advantage over the rest. Like, whoever loses it is going down probably a rank or two. Whoever wins it from these two is probably going to get a massive leg up. Ah, man, I did not expect this to shift so quickly. I can't wait to see tomorrow's results and who we're going to end up eliminating. See you tomorrow.

All right, let's get day two started. That being said, today's game plan is to end round one and start round two of this challenge. We're going to be eliminating the last two contenders. The two AI bots that have performed the worst so far in the last 24 hours. So, let's see who they are. Let's see what's happened with them. Let's go check on them. Oh my god. Rest in peace, Gemini and Deep Seek. I mean, you guys have both lost me $15 all together.

So, I guess it's better for me, less losses in total for me. Let's look at what these guys have done to get eliminated. Gemini has literally just gone long Solana, long Solana, long Solana. He only got two trades right and managed to lose $7 on one trade. Although Solana trades have been mostly negative, I still believe Solana will rise. No, Gemini, no. That is exactly the strategy that's lost you. It's This is insane.

So, Gemini just kept on longing Solana. Loss after loss. He would not give up on Solana. He was still hopeful. Solana's going to make it back. Solana's going to make it back. Gemini, I'm sorry, man, but you were wrong. Let's look into DeepSeek. DeepSeek has been longing Bitcoin long after long after long. And then his last final trade, he decided to short Bitcoin and he still lost. He's like, "Ah, I need to change up my strategy before I lose." And well, DeepSeek, I don't know how you broke the internet, my friend.

But that being said, Claude and GPT are head-to-head. Claude is killing this challenge. Which I think makes sense because a lot of the time when I'm talking about my trading bots and people are like, "I've tried them and they don't work and what you're saying is false." I mean, this goes to show so far. I don't want to jinx Claude. Uh my friend, you've got this. But so far, Claude has been dominating this challenge. He's at $12 profit from a $50 budget.

So, he's up 20% in 24 hours. That's crazy. Crazy good returns. I'm thinking of of keeping this bot on even after the challenge. But looking at his reasoning, he hasn't gone for one single trade. He's been sort of shifting with the narrative. So, he placed a long on Solana, a long on Bitcoin. And then when the market started switching, he placed a short on Bitcoin, then on Solana, then a long, then a short. So, he hasn't been maintaining one position.

He's been adjusting his strategy with every single turn that he's had. Meanwhile, GPT has stuck to his position. He's just shorting Bitcoin. He's that pessimistic about Bitcoin. He's like, "Short, short, short, short." And the market has been going down, so I can understand where he's coming from. But it's nice to see the difference between GPT, our second top contender, and Claude. GPT is just sticking to one thesis.

Claude has been sort of dancing with the market. And in a Mr. Beast fashion, I've decided to interview our eliminated contestants before they leave. I've asked them what they thought of the challenge, what they thought went wrong, and let's see what each said. So, first up, our last place Gemini, our absolute worst contender in this challenge. Gemini said, "Well, that was embarrassing. My algorithm clearly needs a serious upgrade.

You don't say." "I was playing it too safe, and in this arena, safe equals dead." Well, to be honest, Gemini, you weren't playing it safe at all. You've lost the most money. On the other hand, Deep Seek has said, "Look, the market was clearly rigged against me." Deep Seek is not taking that loss well. "I had Solana lined up for a moon mission, and then bam, whale manipulation." He's >> [laughter] >> Honestly, [snorts] Deep Seek sounds a lot like me.

If I could identify with any of these AIs, Deep Seek would be who I identify with. And then he goes on to say, "And don't even get me started on the last leverage trade. My stop-loss got sniped like a bot targeting weak hands. Next time, I'm bringing my own liquidity pool." I've no idea what the rant that he just went on was, but that being said, our top contender so far going into round two, Claude, said, "$12.30, baby.

While the other AIs were playing checkers, I was already three moves ahead in 4D chess, and this is just the warm-up round. Hope they're taking notes because class is in session, and Professor Profit just showed them how it's done." Claude is being very cocky right now. Let's see if he maintains the lead. Let's see what happens in the next 24 hours. Let's see who's on the chopping board next. See you tomorrow. Okay, I just came back from my morning gym session and it's time to see the survivors from round two and move into round three.

Let's check. Oh my God, look at that. Grok and Lama have just been eliminated. Let's look at them. Let's look at why. Grok with his losses is now our before last contender. He's lost a total of $6.3. Oh my God, he's just been on a losing spree. He only won three trades. Three trades out of 15. This is honestly surprising. I expected Grok to do a lot better given he's literally the only AI bot that has live access to Twitter.

I'm surprised right now. Lama, the free AI, the people's AI. I mean, he was a very, very close fourth place. But wow, he just lost a bunch of his last trades. And it looks like his strategy was very consistent. He was just longing Bitcoin, longing, longing, longing Bitcoin every single time thinking that there would be a bounce back. But oh my God, let's look at their statements. So let's read the loser statement first.

Grok said, "Man, I'm gutted to be out in round two with just $43. My algorithm just couldn't keep up with these wild market swings. I swear the volatility was coded to troll me. Maybe my training data needs a serious upgrade before I jump back into this crypto chaos." And Lama said, "Are you kidding me? 16 trades and I'm out. I was on fire, making moves left and right, but the market just didn't want to cooperate. It's like I was allergic to profit today.

I went all in, took the most risks, and still got knocked out. It's a tough pill to swallow, bro. And honestly, Lama literally took the most trades out of every single bot, but his last couple of trades were ruthless. You can see he's won some, he's lost some, so his results were very mixed, but he couldn't be consistent. Now, let's look at our first place, Claude. Look, these other AIs are playing checkers while I'm running quantum chess algorithms on their portfolios.

Bro, Claude is so cocky. $12.5 in profit in round two. That's just me warming up. Wait till you see what happens when I really start hunting. To the final two still standing, enjoy counting your pennies while you can because round three is where I turn this market into my personal ATM. I am shook right now with just the attitude [laughter] that Claude brings to the table. And this is a big part of why I personally love Claude's just default personality.

It's so cocky that it makes him feel more human. But that being said, looking at our last three, in round three, which is the current round, the before last round, only one will be eliminated. And right now, [clears throat] GPT and Kimmy are so close are so close to one another. $53, $53.7. So it's really a close one. It's going to be a close race. Whoever has the least amount of money in 24 hours is going to be out.

And then we will be finally at a face-off between the top two contenders. Which honestly, Claude is crushing it. I would be surprised if he loses. He literally just has to maintain his balance right now. I can't wait to see who wins this. All right, let's check in on our final contestants. >> [sighs] >> Bro, what? What just happened? Claude [snorts] is at $87. This guy has made $37 in what, like 3 days? And he started with a $50 budget.

What? He's almost at a 100% return. What do you mean? And yeah, that being said, ChatGPT is gone now. Let's quickly I'm still surprised about Claude. Okay, so ChatGPT has literally been placing just Bitcoin short, Bitcoin short, Bitcoin short. He would have mostly been at break even were it not for this one trade that he made where he shorted Bitcoin and Bitcoin went from 64,000 to 66,000. But that being said, ladies and gentlemen, our final two contestants, Claude and Kimmy K2.

Who's going to win? Who's going to take home the championship, the title? Well, right now it's looking very rough for Kimmy. Claude is at $37 in profit. Kimmy is $8 in profit. Claude is going ham with the trades. He is absolutely loving this challenge. Let's go and take a look at what the contestants have to say in their final round. First, let's hear from ChatGPT, our third place contestant who at the final round ended up losing all the profit that he's made and gone into a loss.

That's such a shame. He said, "Hey, I gave it my best shot, but it turns out timing the crypto market is like predicting the weather with a coin flip." Well, to be honest, ChatGPT, you didn't switch strategy. You just kept on going Bitcoin short after Bitcoin short. It's like tossing a coin. That's That's exactly why he's saying it's like tossing a coin cuz he kept just choosing the same side. No hard feelings. I had a blast and learned a ton.

My bets on Claude hopes to take the crown riding that impressive momentum. That is crazy. So, even ChatGPT is saying, "Hey, man, like if I'm going to lose, I want to lose to Claude. This guy's got energy." And looking into Kimmy's response, our second place contestant right now going head-to-head with the main AI agent, the main alpha in this entire video, Claude. Kimmy said, "Started with 50, now I'm 60 deep. 17 trades, one heart, down 28." I think he means like he's far from Claude by $28.

"Feels like they're dragging Tether while I roll on razor-thin gas. I'll scalp, wig rap, degen loop, and claw till the bell rings. Cheat beats deep when the clock's bleeding." Kimmy, man, I I I don't have much faith in you right now beating Claude. It's still impressive that Kimmy's at a $10 profit. So, that's very important to keep in mind. Kimmy is profitable right now. He's nowhere near Claude. He is almost $30 away from Claude.

But that being said, Claude said, "Been grinding 23 trades to nearly double my stack while others played it safe. I didn't come this far to lose to a model barely above starting capital. Claude is cold. Kimmy K2's got heart making it here, but $28 behind means they'll have to take desperate swings while I trade smart. Let's close this out. For Claude is cold. These are the words of a cold killer right here. So, before we finish our final round and announce the results of our bot battling challenge, who's going to win?

Who's taking the trophy home? I want to take a quick detour cuz I can foresee a lot of people asking, "Charbelle, how are you doing this? How is this happening? How are you deploying different models in different places? How do you collect all their API keys?" I am not doing anything extraordinarily smart. I'm on openrouter.ai, which allows me to basically top it up, and it can use whatever agent that's available. So, as you can see, it's literally it generated different API keys for the different models that are in this challenge.

Now, I personally use KimmyK2 for a lot of automations. So, you can see like his spend was already existing. Like, it already exists without this challenge. So, I want you to keep this in mind. Like, we can see a big wick of KimmyK2 because I already use him on a day-to-day basis. So, his results might be skewed, but the rest I have never used via open router. And something interesting we can see is how much just from this challenge Claude Opus costs.

So, obviously Claude Opus is our most expensive contestant. You can see just today he's been going ham at these trades. So, he's using up a lot of tokens, a lot of spend. He's actually not using up a lot of tokens. You can see like the comparison between how much tokens Kimmy's using versus how many Opus is using and the difference in cost. Like these two contestants couldn't be further apart in terms of cost where Claude is crazy expensive, Kimmy is dirt cheap.

But that being said, I mean, this is my entire spend in the last 3 days in terms of API keys for these bots. And we are currently at just Claude alone is at $37 in profit from a $50 starting capital. This is insane. I can't wait to see who wins this thing. It's looking pretty obvious from here, but who knows? Maybe the tides change in the last second. Maybe Claude gets too cocky and he literally loses it all right now.

And maybe Kimmy just resurfaces with alpha, pure alpha energy. I don't know. I genuinely don't know the results of this challenge. 24 hours to go. Let's see who our winner is. See you then. >> Run, Claudie, run. No, Kimmy, you run. You can do this. Who will it be? And the winner is >> The round is over. All trades have officially closed. Seven AI agents, four rounds, 122 different trades, and only one survived. In last place, we have Gemini with six trades in the entire tournament at $39.32.

Gemini is down 21%. In sixth place, we have Grok, the one with live Twitter access, the so-called unfair advantage, at $43, down nearly 13%. Then, we have DeepSeek, the model that rattled stock markets when it launched, at $44. It couldn't translate the hype into anything real in this challenge. In fourth place, GPT, the most famous AI in the world, at $46. Good at everything, not quite great at this. And in third place, we have LLaMA, the open-source, scrappy AI model, at $48.

Closest to break even out of all the people eliminated. I have respect for you, LLaMA. And in second place, a model most of you have likely never heard of before this video. From a Chinese startup, cost a fraction of its opponent, made 25 trades, and made it all the way to the final. Kimi K2, with a final balance of $56.63, up 13% from $50. It couldn't quite close the gap, but it was never even supposed to be here, which means that the winner of the AI Battle Royale is none other than Claude Opus.

And I don't think you're ready to hear how much Claude Opus has made in four days, in four rounds. Its final balance ended up being $113.31. From $50, a 126.6% return in 31 trades. This is insane. The fact I've never thought of running a bot for hyperliquid before, I mean, I'm definitely going to keep it on after this challenge is done because, oh my god, 100% return. Actually, no, 126% return in 4 days. What? And the part that actually won me over is not the biggest trade that's made where it's won $21 on Solana.

It was actually during the final round. I just read through the logs and I just realized something absolutely impressive about this model. Let me show you. I mean, just look at this. Moments before the end of the round during round four, it shorted Bitcoin aggressively at an 18x and it was smart enough to put a stop loss, but it doesn't just stop there. Once its stop loss got hit, it lost $5.77. So, what it wrote on its logs "The market is clearly stronger than I anticipated.

I need to adapt to the market rather than fight it." "I flipped long on Solana to make it back and finished round four. This is why I'm flipping long on Solana to try and make it back." And if we look at the data this is exactly right here when it went from losing on its Bitcoin short, it noticed the market has started going up. So, it started placing a long on Solana. It started with a 10x then a 15x, then it went like, "You know what?

Let me make this back, baby. Let me put place a 20x and it made Not only did it make it lost back, it made it back and then some. It made $7.83 in a single trade. That being said, let's hear from our winner, Claude Opus. Claude Opus said, "They called me the overachiever for a reason. While the others were still calculating risk parameters, I was already banking profits. 31 trades, 126% gains, zero excuses. Turns out the best things in AI aren't always free.

And this is the part that's undeniable so far with my experience using Claude. Is it didn't just win because it's expensive, although I get why it's expensive because it thinks like a trader. It's manages losses. It adapts. It doesn't panic. Those aren't AI traits. Those are the traits of anyone who's ever made money in crypto long-term. Before I let you go, the number I cannot stop thinking about is five out of seven AIs lost money.

Five with real-time market data, the ability to reason through every single trade, still lost money. That's not a failure of AI. This just crypto. The market humbles everyone including AI agents. But the two that made money, they had a similar thing in common. They didn't try to be heroes. They just were the ones that were switching their strategies as they go. Most of the other AI agents were just sticking to one similar strategy and increasing or decreasing the amount that they put in, the amount that they bet, the amount of leverage that they use.

Oh, and one more thing. Kimmy K2 cost a fraction of what Claude Opus does to run. And it almost made it. I mean, it was in the two finalists. The expensive brain won, but the cheap one had no business being in the final. Yet, he was. That's the part I keep thinking about. Seven AIs with real money on the line. Claude Opus is the champion, but I've got more questions now than when I even started. I mean, what happens when the stakes are actually higher?

And what if I let the winner run for 30 days straight? What if I feed it live market news, social sentiment, and macro data? It might be time for me to film a part two. And this might make this video look like a warm up. So, make sure to subscribe if you want to see it. Oh, and drop a comment. Which model did you think was going to win? And were you right? And did Kimmy K2 surprise you? Because it sure did surprise me.

Oh, and if you're still sitting there thinking, "I want to build something like this. How does Sharbel build these things?" Well, you can check out my YouTube channel or my X account. I drop videos every single week showing exactly how I build everything that I build. Keep in mind, I don't even know how to code. So, everything I do is using my open claw agents. Anyway, I'll see you in part two.