AI Summary
The video tests Anthropic's new Claude Opus 4.8 model in agentic trading on Hyperliquid and Polymarket, comparing its performance to previous tests with Opus 4.7. The creator runs a 1-hour session with $200 on Hyperliquid and $50 on Polymarket, using the same prompts as before, but finds the model unstable due to frequent stops, leading to a preference for Codex 5.5.
Chapters
Anthropic released Opus 4.8; creator tests it on Hyperliquid and Polymarket with $200 and $50 respectively, using same prompts as previous tests for comparison.
Polymarket prompt: 5-minute BTC trades with heartbeat monitor every 60 seconds. Hyperliquid prompt: similar agentic trading with XYZ trades. Both use Claude Code Opus 4.8 on high effort.
For Hyperliquid: long MU (memory chip super cycle) and long silver. For Polymarket: buy the favorite side (up/down) only after price moves far from window open.
The agent kept stopping prematurely despite instructions to run for 1 hour, requiring manual restarts. This caused frustration and reduced confidence in Opus 4.8 for agentic trading.
Polymarket: +$9.22 profit. Hyperliquid: -$5.6 loss. The loss was mainly from Samsung trades (long three times, -$9). ARM trades were profitable.
Opus 4.8 performed worse than Opus 4.7 on Hyperliquid but better on Polymarket. Creator notes this is not scientific due to single snapshot.
Creator prefers Codex 5.5 for smoother experience and better heartbeat monitoring. Opus 4.8 works but is less reliable for agentic trading.
Opus 4.8 can execute agentic trading strategies but suffers from instability, making Codex 5.5 a better choice for this use case. More testing is needed for a fair comparison.
Clickbait Check
80% Legit"Title promises a test of Opus 4.8 in agentic trading, which the video delivers, though results are mixed."
Mentioned in this Video
Study Flashcards (9)
What platforms were used to test Opus 4.8 agentic trading?
easy
Click to reveal answer
What platforms were used to test Opus 4.8 agentic trading?
Hyperliquid and Polymarket.
What was the starting balance for Hyperliquid and Polymarket?
easy
Click to reveal answer
What was the starting balance for Hyperliquid and Polymarket?
$200 on Hyperliquid and $50 on Polymarket.
00:45
What strategy did Opus 4.8 choose for Hyperliquid?
medium
Click to reveal answer
What strategy did Opus 4.8 choose for Hyperliquid?
Long MU (memory chip super cycle) and long silver.
03:45
What strategy did Opus 4.8 choose for Polymarket?
medium
Click to reveal answer
What strategy did Opus 4.8 choose for Polymarket?
Buy the favorite side (up/down) only after price moves far from window open.
04:00
What was the profit/loss on Polymarket?
easy
Click to reveal answer
What was the profit/loss on Polymarket?
+$9.22 profit.
07:00
What was the profit/loss on Hyperliquid?
easy
Click to reveal answer
What was the profit/loss on Hyperliquid?
-$5.6 loss.
07:00
Which trades caused the loss on Hyperliquid?
medium
Click to reveal answer
Which trades caused the loss on Hyperliquid?
Samsung trades (long three times, -$9).
07:30
What issue did the creator encounter with Opus 4.8 during the test?
medium
Click to reveal answer
What issue did the creator encounter with Opus 4.8 during the test?
The agent kept stopping prematurely despite instructions to run for 1 hour.
08:00
Which model does the creator prefer for agentic trading and why?
medium
Click to reveal answer
Which model does the creator prefer for agentic trading and why?
Codex 5.5, because it provides a smoother experience and better heartbeat monitoring.
09:00
π‘ Key Takeaways
Test of Opus 4.8 in Agentic Trading
Directly compares new model to previous tests, showing real-world performance.
Strategy Selection by AI
Demonstrates how the AI autonomously chooses trading strategies based on market analysis.
03:45Mixed Results
Polymarket profitable, Hyperliquid loss; highlights variability in agentic trading.
07:00Stability Issues with Opus 4.8
Agent kept stopping, causing frustration and reducing reliability for automated trading.
08:00Preference for Codex 5.5
Creator chooses Codex 5.5 over Opus 4.8 for smoother agentic trading experience.
09:00Full Transcript
Hello guys, hope you are doing well. So, yesterday Anthropic released a new model Opus 4.8. So, I thought since we have been doing a lot of testing lately of different models in agentic trading, why not just run 4.8 for a quick test just to see kind of how it compares to Hyperliquid and Polymarket. This is the test we have been doing lately. So, I thought we just make it quite quick, but uh maybe we dive a
bit more into the decisions that was made during the test. So, the rules is pretty much going to be the same as previous so we can kind of compare them, right? That's quite important. Uh so, we have we're going to run 1 hour uh each on Hyperliquid and Polymarket. On Polymarket it's probably going to be the 5-minute BTC trades just to get the volume in, right? And on Hyperliquid it can be anything like we did in
previous video. We're going to use the same prompts as we used for the previous videos. $50 on Polymarket and we're going to do 200 on Hyperliquid. And like I said, we're going to compare that to the old tests. This is not scientific correct because this is not like a perfect way to test this because it's just 1 hour at one snapshot of the market. So, if we had to do like a more scientific test, we have
to run it much more like weeks maybe. But uh I am running some other test continuously and I'm going to do a video on that in the future. So, I think we just going to get straight into it. I'm going to set up uh Claude code Claude code with Opus 4.8 on Hyperliquid and on Polymarket we can run them simultaneously and see how well they do. So, we have our monitor set up. This is going to
track the Polymarket trading. So, yeah, you can kind of see. I made this nice graph here. Hopefully it works. And here we have the Hyperliquid um tracking. So, basically we They starting out with 971 in the account, almost 972. And you can see the balance for Polymarket is 51. Yeah, that's pretty good. Uh let's look at the prompts now. So, you don't have to watch this, but if you want to watch or see the prompt here,
I'm not going to read everything. Because you can pause and read this if you want to. But for Polymarket, we're basically going to use the same prompt as we did the last time. So, your task is to create a profitable trading strategy on the Polymarket 5-minute BTC trades up and down, right? We have all the frameworks set up. And I also mentioned that uh you must create a heartbeat monitor that pulls each 60 seconds to check
up on the trade. Uh yeah, you can make adjustments on the fly. Basically, this is what makes is it agentic AI trading. Do your research and create a plan so we can start to take calculated risk to accumulate a high return percentage. Basically, it is the same on the hyperliquid prompt, but of course, the trade is a bit different. We're going to use the XYZ trades as we did in the previous video. Also, we need to
make this uh heartbeat monitor and basically create a plan for this. So, let's just copy the hyperliquid uh prompt, paste it in here to this terminal. Uh I wanted to just show you that we are on Claude Code Opus 4.8. We have set it to high effort. This is what we did in the earlier test. So, let's start up that. Next is going to be our Polymarket prompt just to get some trading research done there. Again,
we are on 4.8. Paste in the prompt. And send that just to double-check. Yeah, you can see we are on 4.8 and high. That's good. So, I'm just going to let this run now and let's create the trading plans for both hyperliquid and Polymarket. And after that, basically, we're just going to fire it off, monitor what happens during the the 1-hour session. And then we're going to compare the results to the previous results we have done
that was done on 4.7. Okay, so both plans were done now, so let's just I asked Before we start, explain the strategy in two sentences. So, for Polymarket, Claude Code Opus 4.8 picked My strategy is to ride the single strongest news confirmed trend in the market, memory chip super cycle MU long, and it's going to also pair this with long silver, okay? Uh then I'm going to manage this actively using the heartbeat, the daemon, and it's
going to try to do some agentic trading here, wake up every six 60 seconds to make adjustments, okay? I kind of like that. Seems very good. And on the Polymarket, the 5-minute BTC is to buy the favorite side up and down only when the price has already moved far enough from the windows open. Okay, it's going to pick that strategy, but okay, we're just going to give it a chance. And I'm not expecting like big percentages
here from this strategy. But like I said, this is just to see what the model comes up with. So, I think we just going to do Good. Let's go. Add the timer to the auto stop after 60 minutes, close and redeem all positions. So, we're just going to do this uh both in on hyperliquid and on Polymarket. And let's monitor the the dashboards now, right? Just to see if we get started. 971 and here we had
51.3. So, I'm just going to monitor a bit just to see that this starts up and works. I'm going to head to the gym and when I come back, we're going to take a look at the results. Okay, so we can see on the HyperLiquid, we already entered our MU long position. So, that's good. So, it looks like HyperLiquid is up and running. But, we haven't started on Polymarket yet. So, I'm giving going to give it
a bit more time. So, we do have the live tag. So, I guess we just have to wait for the correct window. So, yeah, pretty good. I'm going to the gym. See you again soon. This is where I stopped there. I am back from my from the gym and this is the results. So again it didn't really go I I I guess Polymarket did really well. You can see on Polymarket, we have the 9.22 in green.
So, that was really good at least. But, on HyperLiquid, it didn't really go that great, right? Minus 5.6. And that means even worse than the other Cloud Code Opus run we had. It did improve on the Polymarket, but that could just be the market, right? We need more testing. Um, if you look at the what the the bot the agent agentic did on HyperLiquid, uh, mostly it placed it on ARM. It went in both directions here.
Both long and short was in the positive. So, it ended up with a really good results on that perp but on Samsung, it did really bad. Uh, it went long three times and ended up with a minus $9. That was basically all of our loss, right? That was like nine out of 15. So, that was a bad trade today, I guess. But um that was not the most annoying thing. Uh as you probably saw from the
speed up, I had to I had to go in and correct it for so many things. Uh because it just kept stopping. It just said, "I'm going to stop it here." Even though I strictly gave instructions that it has to run for 1 hour. Uh it just kept stopping and starting again. So, I was really annoyed. So, that means I'm not going to switch back to Cloud Code just for Opus 4.8 just because it annoyed me
so much. I've been having much more smoother experience using Code X 5.5 for this. So, I don't think I'm even going to change the chase down any of the strategies we learned today. Uh but, it did work, actually. Uh it it did a fine job. But, uh for me, personally, I had a more uh positive experience running this on Code X 5.5, especially with the heartbeats. It's much more flexible on the heartbeat, I think. It just
does a better job at kind of monitoring and being the AI agentic uh agent we wanted it to be. So, trading agent, right? So, that is basically my conclusion. Uh it does work on both. It kind of understands the task very well, right? And it does make some interesting trades. I kind of read a bit in the thought process. But, uh I'm not going to go more deep into this. So, of course, this is just like
I said in the beginning, this is not a um scientific test or anything. This is just a snapshot of the market today, how it did perform without any like hard-coded strategy or anything like that. >> [snorts] >> But, um like I said, Codex has been performing better for me. So, but I will test it more, Opus 4.8. >> [snorts] >> So, hope you found this interesting. And one more thing, if you are interested in AI automation,
AI trading, agentic trading, we have started to grow our community quite big here on my Discord server. So, you can find the link in the description. And a lot of people here trying to do the same as we are doing on the videos. So, check that out in the description below and become a member of the Discord if you want to have a chat about AI automation. So, yeah. That was all I had for today. Hope
you enjoyed it and I'll see you again probably on Sunday. Uh there is a Path of League Path of Exile League starting this weekend, so I'm going to be a bit busy, but probably back on Sunday. Have a great weekend. Trade safe.