--- title: 'Testing Claude Fable 5 for Automated Trading: 71% Win Rate in 15 Hours' source: 'https://youtube.com/watch?v=DqxUGfa6VfQ' video_id: 'DqxUGfa6VfQ' date: 2026-07-14 duration_sec: 0 --- # Testing Claude Fable 5 for Automated Trading: 71% Win Rate in 15 Hours > Source: [Testing Claude Fable 5 for Automated Trading: 71% Win Rate in 15 Hours](https://youtube.com/watch?v=DqxUGfa6VfQ) ## Summary The video demonstrates testing Anthropic's Claude Fable 5 model for agentic AI trading on Polymarket's 5-minute up/down market. The model analyzed 24 hours of data, developed a strategy incorporating fees and edge calculations, and executed trades autonomously, achieving a 71% win rate and $82 profit in 15 hours. Key innovations include a cron job for periodic strategy review and a 'deep long shots fading a jump' approach for high-payoff trades. ### Key Points - **Initial Results** [00:00] — After 10 hours of trading, the bot achieved a 71% win rate and $41 profit, later increasing to $82 profit over 15 hours. - **Data Collection** [00:30] — Collected 24 hours of snapshot data from Polymarket's 5-minute up/down market just before the model release, used as input for strategy development. - **Prompt and Strategy Development** [01:30] — Prompted Fable 5 to analyze data and find a plus EV strategy. It produced a formula: buy when fair value exceeds executable ask plus fee, with specific conditions like edge persistence and position sizing. - **Cron Job for Monitoring** [03:30] — Set up a cron job in Cloud Code to wake every 2 hours, analyze trades, and adjust strategy if needed. The model performed health checks and made no changes during the night. - **Deep Long Shots Strategy** [05:00] — The model identified a 'deep long shots fading a jump' trade: low probability (22.7%) but high payoff ($59 win vs $13 loss), yielding a 4.5 cent edge. This was a novel approach not seen before. - **Model Switching to Save Tokens** [07:00] — To conserve tokens, the user switched to Sonnet for simple queries and used Fable 5 only for complex analysis, demonstrating cost management. - **Interactive Website Creation** [08:30] — Fable 5 created an interactive HTML site explaining the strategy, including the formula, live decision engine, and trade history. ### Conclusion Claude Fable 5 shows strong potential for agentic AI trading, with innovative strategies like deep long shots and autonomous monitoring. However, token consumption is high, so model switching is recommended for cost efficiency. ## Transcript So, pretty exciting news if you ask me. Uh Anthropic just released their Claude Fable 5 model and I've been testing it on some agentic AI trading over this night, at least for me it was over the night. And I've been pretty impressed by the results. You can see here, barely if you can see, we are like plus $41 after running this for like 10 hours. 71% win rate. I guess I can show you here. So, we are like 71% win rate. We are 41 plus up, right? And you can kind of see here. But, what was mostly interesting was the strategy it picked. You can also see here 1 day, uh $82 up on our account. So, it's been really good, to be honest. And that kind of surprised me a bit. But, we're going to go more into the strategy Fable 5 picked for us anyway, because like I said, it was pretty interesting. But, the way I set this up was uh I had a bit of luck. I just collected like 24 hours of data from Polymarket. Uh I just wanted to do a new run on the 5-minute up and down market. So, I collected some data just a few hours before the release. I didn't even know that. So, that was pretty good timing. And the the workflow was pretty standard uh as I do with all new models. Basically, I took the data, I pointed the Fable at the data, right? To analyze it, come up with a strategy to be like plus expected value. Build it and execute it. But, I did one thing that was a bit different now uh that I'm going to show you that is kind of important if you do agentic AI trading. So, I think we're just going to get into it, look a bit about how uh Fable 5 got to get used to that word, set this up. And what kind of changes I haven't really seen before in these kind of uh setups. So, let's just have a look and I'm going to kind of walk you through how I set it up and kind of the results. So, if we head over to Cloud Code here, you can kind of see the Yeah, I just want to scroll all the way up here if I can. Fable fires here, a new model for complex running work. Perfect. So, I just want to go through the simple prompt I gave this because you can see it's a bit scattered here, but basically I just tagged my data files. I guess I can show you that. So, here you can see all the data I collected beforehand. So, there are quite a lot of data points here. Yeah, I guess you can see it here. Here you can see it, some of it, the data we collected. Here's a bit more. And basically it's just a bunch of snapshots from from a longer session on Polymarket, 5-minute up and down. And you can kind of see I tagged that. I said we have 5 hours. This was the live data I had collected, not the 24 hours. But basically I pointed it at that anyway. And you can see from the 5-minute up and down market from Polymarket, your task will be to analyze the data, do test, and the best signal strategy for a plus EV trading bot setup on this market. This is the way I use when I test all models. Think hard, write the test you will need to find the best strategy from the data to run live. Use your 100x financial genius brain to solve this. I just wanted to add this kind of fun to see if we got a response, but we didn't really get any funny response from that. It just started straight into planning out this. And you can see Yeah, it kind of went through all the markets here. I'm not going to go through every single step here, but I spent a substantial amount of time analyzing the data. And it did something I haven't really seen like in detail before. It kind of took like one on one snapshot and did some testing on those. And yeah, it was kind of exciting to watch actually. And you can see we kind of ended up with this report here and a recommended strategy. So, the description was buy the side whose models fair value exceed executable ask plus fee that buy data Binance leads poly market book. Okay. So, we came up with I like that included the fee because that's really important if you're going to calculate like forward earnings and profits because we need to Yeah, take account for the fee. And it created this formula here for uh fair value. And it just set the signal when we're going to buy this. But, what was really interesting, I'm going to show you here a bit further down. So, what we kind of ended up with was this recommended live setup here. Buy Buy when fair minus ask minus fee is equal to 0.04. Only 15 to 180 into the window. Edge must persist two consecutive checks. Uh fill on kill at ask. One trade per market. Hold to resolution. No stop loss. 5% of bank roll per trade. One fourth Kelly with staleness volume receipt guards and 10% daily loss halt. At observed depth, that's roughly three to six trades per hour. And that's kind of what we saw. But, I'm going to show you one thing I added because this is I think it's really smart when you're doing this identic trading here. So, here you can see I said, "We've been trading for 3 hours now. Analyze the trade so far. Make adjustments if you see any to improve the uh expected value. If not, decide what to do. Work hard. Think like a genius." So, I just added this and it kind of went over, did some adjustments based on the trade we have done so far. I saw a bit of an uptick after we did this. Uh so, you can see here we had something that I found was really interesting. And that was something called deep long shots fading a jump. So, that was something I have not seen before. And I just want to read what he kind of said about this. And it talks about it here, so you can see this is a trade that would probably lose. The model itself only gives it 22.7%. That is not a malfunction or a bad entry. It's plus expected value because the $59 wins dwarf the $13 loss. Break even is 18%. Model says 23%. Hence the plus a 4.5 cent edge. So, I have not seen that before actually in my analysis and I've done a lot of this. And I found it interesting that he kind of split up his strategy to just sometimes do these more conservative trades, but throw in these long fades to try to hit more of like a long shot that is kind of a high payoff. So, this is something I implemented, but I also did something that I thought it was really smart and something you should look at when you're setting up your algorithmic trade. So, let me just go a bit bit further down here. So, here is what I did because I needed to go to bed right, but I still wanted this to run over the night. So, after I have the model has done some adjustments here over the yeah, the start of the trade, the three first hours. But then I said, "Can you set up a monitor that wakes up every 2 hours, checks on the trade we have done, make adjustments if needed, and restarts the bot?" So, what we did, basically what you can do in Cloud Code is set up these cron jobs, right? So, for every 2 hours while I was sleeping, it kind of woke up, took a look at what we have been doing so far. You can see here each wake up call I will do a health check, analyze all the trades, check up on the the formula or the strategy, adjust on strong evidence, restarts when warranted. So, if we scroll further down here, you can see the first schedule review was bought healthy, no changes warranted. So, that was pretty good and we kind of got a update here. Same on schedule review two, no changes. Again, we kind of got into the fifth review. Session is now very positive, no changes. And I just said, let's take a pause here because uh yeah, I wanted to do the video. So, so far it's been really interesting. And I just wanted to go quickly back here and check some of the interesting positions took here. So, someone that kind of stands out, let's say this one. See See the span here. It took a position on 94, 12 shares, and won 70 cents. That's a really strange position. Risk 13, 14 I guess, to win 70. But also down here, you can see we entered with 29 shares at 38, and we won 17. So, it has a really big span here. You can see we also entered with 29 at 42. So, can we see the first one? I guess we can't because we made all our trades. But basically, the span of the the positions is has been taking has been really interesting. And I'm going to continue running this because I feel it's just been running for like 15 hours, and that is not enough. But like I said, we have been doing really good, like almost $100 profit so far. So, it's going to be really interesting to see where this goes. So, uh I wanted to show you a couple of things about the Fable model because it's really really uh expensive or like token hungry. So, let's say we do we go into our cloud code here. You can see we are set up with Fable 5. So, what I did yesterday, instead of just running this, so let's say I wanted to check up on our data. Uh I'm going to do model switching from now because it's just going to drain my uh my account if not. So, I hear I'm going to switch to Sonnet. I'm going to see uh say something like find where our data is uh stored. So, here I'm just going to run that on Sonnet, right? Because uh yeah, I want to save some of those precious tokens. Okay, so you can see that was pretty quick. So, now we have a a summary of where all the data lives. Let's switch now. Let's do model uh let's go back to Fable. Okay. Switch to Fable. We can do slash effort. Uh let's say to extra high. I'm not going to do ultra code. Uh so, let's do extra high because we're going to do a simple analysis of the trading we've been doing for so far. And we can ask Fable to kind of explain our strategy so far. Maybe we can get it a bit more um clear that what I've been doing. So, I just asked it to create an interactive stunning HTML site that explains the strategy. Dark team. Also include how it calculates the signal to buy. But, there's one there's one interesting caveat here, right? Because in the beginning on this long shots, it placed like 80 shares on point 15 or something. So, if we have uh tightening the cheap down entries, uh which would have turned the ledger into plus 41 into a roughly 120 up. So, we did actually make a a bit of a a mistake in the beginning here, but uh it had corrected for that over time uh during uh some um questions I asked it. So, it seems a bit more tight on that now. Uh I haven't seen it enter those low low uh entries now. But, uh let's see how it goes with the website if we can see a bit more how it decides to actually um to go into a trade here. Okay, so we got our website here, so you can see uh be right before the market re-prices. That is kind of the headline. So, here is the edge Binance leads Polymarket lags. So, I think it has a Binance uh websocket connected up to this. So, you can see it kind of tries to explain uh here. And one formula for gates, we have this formula here that it created. So, this is kind of where it decides where to go in. So, it follows this formula here. So, yeah, it kind of bugged up a bit here, but this is kind of explaining uh how the formula works. Okay, that was pretty cool. And try it, the live decision engine. So, we can set uh BTC has moved uh 12 bps, that is a buy up. And we can adjust the elapsed in window because we need to be before 120, was it? Something like that. Volatility, we can adjust that. Uh up and down, down and up, okay. That's pretty cool. Uh what else do we have here? The edge is real, it's also loop-sided. Okay. So, we can kind of see uh all the trades we have been doing. And here are kind of our bad trades. This was the one we did be under 20 or 50. Okay. And yeah. All right, that was pretty cool. So, it did create this interactive website of how we've been doing so far. Okay. So, yeah. Uh I think that's kind of my initial testing of Fable. And I am going to go back now. I think we just going to go back here, and I'm going to say uh restart trading because I want to continue keep on testing this. So, I'm just going to start up the bot again. So, if you want to kind of follow along to see how it's going to do, I'm going to keep updating this on our Discord. You can find that in the description below. If you want to become a member, I'm going to keep updating how this setup is going to do over time. Uh at least I'm going to try to do that. So, you can see we are restarting this now and you can see we are live again. So, that's pretty good. So, hopefully this is going to continue, but I think it'd be kind of going to reverse a bit down towards the mean, but uh hopefully over time we can at least be a bit in the green. That would be interesting. So, yeah. Uh definitely go check out Fable, but uh like I said, be a bit mindful when you use it. Like do some model switching and stuff like that because it's really going to eat up your tokens. But, so far I've been super impressed and I really like the way it uh uh it came to this setup here. So, uh very exciting and I'm going to keep testing this. So, if you want to more information, more content like this, give this video a like and subscribe and I'll see you again hopefully in a few days.