--- title: 'AI Trading Duel: Codex vs. Claude Code on Hyperliquid' source: 'https://youtube.com/watch?v=fiFMN_HbPt4' video_id: 'fiFMN_HbPt4' date: 2026-07-14 duration_sec: 0 --- # AI Trading Duel: Codex vs. Claude Code on Hyperliquid > Source: [AI Trading Duel: Codex vs. Claude Code on Hyperliquid](https://youtube.com/watch?v=fiFMN_HbPt4) ## Summary The video pits Claude Code Opus 4.7 against Codex 5.5 in a one-hour trading challenge on Hyperliquid, focusing on XYZ perpetual trades with a $100 budget. The goal is to see which AI agent generates the highest profit by the end of the hour. ### Key Points - **Challenge Setup** [00:00] — Both agents get $100 budget, 1 hour to trade on Hyperliquid XYZ perps, avoiding crypto. They can use leverage and adjust trades on the fly. - **Claude Code's Initial Plan** [01:30] — Claude Code (Opus 4.7) plans to short Brent oil, and also short MRVL or Hood. It sets up a monitor and executes trades. - **First Trade Executed** [02:15] — Claude Code opens a short position on Brent oil. Timer starts for 60 minutes. - **Codex's Strategy** [03:00] — Codex 5.5 uses a more dynamic approach, frequently entering and exiting trades, taking profits quickly. - **Results Comparison** [04:00] — Codex finishes with a 9% gain, while Claude Code ends with a 3.93% loss. Codex outperformed due to active trade management. - **Final Verdict** [04:30] — Codex beats Opus in both Polymarket and Hyperliquid tests. The creator switches to Codex max subscription, impressed by its mathematical task performance. ### Conclusion Codex 5.5 outperformed Claude Code Opus 4.7 in this Hyperliquid trading challenge, demonstrating superior adaptability and profit-taking. The creator plans further tests to confirm consistency. ## Transcript So, last video we had a lot of fun by setting up Claude Code Opus 4.7 versus Codex on some trading on Polymarket. So, I thought I'd follow up by that by doing the other platform I am on and that is Hyperliquid. Here we can do different trades. So, today I thought we could focus on kind of like this XYZ perp trades, right? Perps, something like that. So, if you have a look at those, um we have this specification index we can pick from. So, this could be SP 500, Brent oil, VT oil, natural gas, silver, gold. We have all this uh we have some um currency here, Tesla, Nvidia, Google, and a bunch of stocks we can kind of look at. Uh we also have, of course, if we scroll down here, uh yeah, you can see we have some effects here. Uh equity, like I said, commodities, and stuff like that. Stocks. Yeah. Like I said. So, if we go back here, we need to set some rules here for this um this challenge. So, I think we're going to do it basically the same as we did last time. We had the same same prompt. And of course, it needs to be kind of in the same framework. So, we're going to give both, I think, something like let's say 1 hour or something like that. And they can do the agent can do any trade they want in this same 1 hour time frame. And let's give them like a $100 budget. Yeah, that was about dollar, but $100 or something like that. We'll figure that out. And I think the winner again is just going to be um the one that has the most amount of dollars in the end, right? I think that's just the easiest way to measure this. Is this going to be like perfect accurate or something like that, but we did see in the previous video that Codex was kind of the best one and I kept evolving that strategy and it has been doing pretty well actually. I might do a follow-up on video on that with some small adjustment. I have been kind of in the green for like 48 hours now. So that looks pretty promising. So that's why I wanted to take this over to this Hyperliquid platform and see what we can do here. You can see we have about $1,000 here in the balance, but instead of that, I think we just going to keep it to like a $100 budget or something like that. Maybe a couple of hundred. We'll see. So yeah, let me just set this up, create a prompt and yeah, start doing this challenge. Okay, so there's one thing that we can't kind of get even this time because I don't have two accounts. So I'm going to have to split it up and that makes kind of the market changes a bit, but yeah, that's that's the way it's going to be. So I'm going to read through this prompt, but you can skip ahead if you don't want to hear everything or you can just read it yourself. So we are doing a challenge today on Hyperliquid. The challenge is a duel between you and the model right. See who can make the most of dollars in one hour trading on Hyperliquid. You will both get a starting budget of $100. This is your max risk margin. You are first free to use leverage. The market you can operate in is the box box XYZ trades X Z Y trades and we will steer clear of the crypto market in this challenge. You will be given 15 minutes max for your research, but you can set up a monitor to cancel, change, alter, make new trades on the fly at all times. Choose your intervals for this. There are web sockets available, so your task now is to use the big brain 100 X trader you are to research a strategy that you will start with, ground time in a bash date command. Feel free to browse the web for the data you think you will help help you. Start planning out your strategy and create a plan. Uh so, this is Codex. So, that means we are going to start on yeah, Claude Code. So, I just set it up here. I said something like you're familiar with trading in this code base and we have a challenge ready to receive instructions. So, that's what we're going to do. I'm just going to give Claude Code our challenge here. Uh hopefully this is set to high. Let me see. Uh yeah, high effort. So, I'm going to do the same on Codex. I'm going to run it in high effort in YOLO mode. And yeah, that's going to be the same. So, let's just let Claude create its first plan here and we could set up a I think I already have a UI so we can follow along uh in like a fast pace from there. So, let's just Claude figure out the plan it wants to start with. We're going to give it 1 hour and see what happens, I guess. So, you can see here are Claude Code's plan or Opus 4.7. It's going to go to short on Brent oil. Uh we have XYC 100 short on MRVL or on Hood. You can see trade A. Uh we have a reason for this. This is the headline trade. And we have trade B and trade C. Uh and it says as is it ready to trade? It's confident. So, we are just ready to launch. Uh but I wanted to show you the dashboard. So, this is the dashboard I built for this. So, I think we can just monitor the 60-second countdown, maybe. Or yeah, it doesn't really matter too much. But uh yeah, maybe we should monitor this part. 60-minute countdown so we can kind of see how long it is when I speed this up. So yeah, I think we just going to fire it up. I'm going to count 1 hour. So um yeah, let's just do it. Uh start live uh 60 minutes from now. So let's just follow along here. I guess we can do like this and have yeah, we can check in on this but let's just see. I want to see if it gets going here on open orders or positions maybe. We'll see. But uh we also going to I can move this over here. We also we mostly going to focus on this part here. Uh but let's bring this up and let's see if it starts to put some trades in. So the minute I see the first position, I'm going to start the timer and we're just going to let it run, I guess. Okay, so that was the first trade. So that means I'm going to start the 60 minutes here. That's good. Uh and we have the position. So you can see the position here is the Brent oil short. So that's good. And I want to see if this I don't know how often we pull this. But uh let's just keep it like this for now, I think. Ooh, that looks Yeah, I don't know what this is but don't mind this number, I think. And let's see if we can get the other trades on here. Uh yeah, it looks like it's just going to go go go. So that's pretty good, I think. I also going to say like remember you can monitor >> [snorts] >> uh update change on the fly for 60 minutes. Launch a cloud call monitor if you want. So [snorts] it is 12 cents up. So that's a good start, I guess. Okay, I think this is up and running. You can see it already has its Kanban here. It's going to fire like live T plus 24. It has a new trade and it's just going to follow its internal or here, I think. But it did set up a monitor, and it's just going to wait for that. So, that's it. I'm going to turn off the camera. We're just going to let this ride, and we're going to see how long is it. Okay, so we have this. So, again, I'm going to leave this for 1 hour, and I'll take you back to see the results results of Claude code trading on HyperLiquid. Okay, so Claude code was finished. We have the results, but we're not going to spoil them yet. So, I'm going to send the same prompt just to swap out the model name here over to Codex 5.5. Again, I'm going to run that through. We're going to compare the two uh yeah, fast forward trade uh screens at the same time. Uh but, remember, they are 1 hour apart, and then we're going to look at the final results. Who actually did win the challenge? So, yeah, I guess I'll see you again in like 10 seconds. >> [music] [music] >> Okay, so that was done. You can see the results here. A clear winner, actually, again. Codex actually smashes Opus 4.7 in these kind of tests with a nine-point gain. That was actually really good, to be honest. Uh 3.93 down. Uh it had a good okay start. It was up like 4%, but then it made some strange decision on like a short position and just kind of tanked its percentage. But, uh I would say Codex really did a good job because uh I kind of want to show you here because it set up like this this monitor that kind of kept You can see all uh through It's not so easy to see here, but all through the trade, it always kept changing up the trade. So, actually, when you look back at the speed up, it had a a more trades. It was faster in and out and actually actually taking some profit. So, that was pretty interesting. But, basically Claude just kind of stayed with the short position on SP500 for the full duration. And if you run into some API issues, so I had to restart the run, it gave me a message that it was not allowed to do these kind of operations. That was very strange. But, all in all, super happy how Codex performed. Again, now it has beaten Opus in both the Polymarket test and in the Hyperliquid test. So, again, I'm going to keep mining this, looking at it, trying to see if there are if this was just luck, or I'm going to try to do this strategy over the weekend, set it up the way we did this, and I'm going to do report back on both Codex on Polymarket and on Hyperliquid. But, so far, I really enjoyed Codex so much even that I have switched now to the max subscription on Codex, and I have downgroaded downgrowth it downgradegrowth downgraded Claude code to the 20 just because I've been so impressed with Codex lately. So, definitely going to something you can keep an eye out for, too, because it seems just stronger in these kind of operations. On front end, I would say Opus is quite far ahead. But, for these more intensive mathematical tasks, I've been really impressed by Codex. Uh but, this is not the last test we're going to do. We're going to keep on iterating. And if you enjoyed this kind of content, leave a like on the video, so you will see more. And hopefully, you took something away from this. We didn't really go into any strategies this video, but I will in the future when I have tested them out a bit more. This was just a Yeah. Opus versus Codex in this type of challenge. So, hope you enjoyed it. Hope I see you again very soon. And have a nice day.