YT YouTube2Text

YouTube2Text → Transcripts → Qwen3 Coder tested: Is it good for algo trading strategies?

Qwen3 Coder tested: Is it good for algo trading strategies? — Transcript

by Algo-trading with Saleh · 4,932 words · 695 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Hey guys, it's S. In this video, we're
0:02going to review the Quen Tree Coder
0:04model. It's the latest open source model
0:06and according to their creator Alibaba,
0:09it is on par with the Clawson Sonnet
0:12model. And considering the cost of this
0:13model is significantly lower. If that's
0:15true, then it's an amazing choice for
0:17us. But as always, what we care about as
0:19algo traders is the performance of these
0:21models for tasks such as writing algo
0:24trading strategies from scratch,
0:25improving our existing strategies,
0:27debugging them, and whether or not the
0:29model has some trading knowledge. So in
0:31this video, we're going to test the
0:32model to see how it compares to the
0:34previous models that we have tested. If
0:36that all sounds good, let's get right
0:37into it. Now, this is the blog post
0:40released by the Alibaba. And as you can
0:42see, it is either beating all the
0:43previously released models or it is on
0:45par with them. And by the way, they're
0:47also comparing it to Kimik K2 model,
0:50which just before this model was
0:52claiming to beat all the other models.
0:53And in fact, I already made a video
0:55about it. It's still being edited by my
0:57editor. But this video, I'm going to
0:59edit myself. So I have no idea which one
1:02you're going to see first. But in either
1:03way, with this type of testing that
1:05we're going to do today and give the
1:07model some ranking, you'll be able to
1:09compare the two very fast. They're also
1:11showing this beautiful chart here which
1:13we can see very quickly which model has
1:16good performance based on their size. So
1:20basically the lower the size of the
1:21models so such as these as you can see
1:24are even smaller than Kimik 2 it means
1:26they'll be cheaper or faster. Although
1:28to be honest in this case the coin 3
1:31isn't cheaper than Kim K2 but it may be
1:35faster but comparing to other models
1:37such as Gemini 2.5 Pro or Cloud Sonnet.
1:40Yes, it's definitely cheaper. Now, even
1:43though this model is open source, it
1:44doesn't really mean that you can host it
1:46on your own local machine cuz taking a
1:49look at the size of the model, it has
1:51480 billion parameters. That means
1:54forget about hosting it on any single
1:56GPU or a MacBook machine like mine. So,
1:58the only thing that you can do is to run
2:00it on a cluster. So unless you are a
2:02business and you have a cluster of GPUs
2:05in your hands, then the only option that
2:07you can use this model is by using their
2:09API with services such as open router or
2:12you can also use Jesse GPT which is our
2:14cloud-based service. Now in this video
2:16I'm going to use Jesse GPT as always.
2:18But if you guys have some privacy
2:19concerns or you don't want to pay us
2:21anything and you want to self host
2:23everything by yourself, then go ahead
2:25and watch this video of mine from before
2:27where I show you how you can use open
2:29router and JAN AI to self-host these
2:32things on your machine. Now when I say
2:33self-host, I don't mean actually running
2:35the model. You will still be using some
2:37API from another service. So with that
2:39in mind, let's get to the first
2:41question. Write a golden cross strategy
2:43that uses the 50 and 100 moving
2:45averages. close the position when the
2:46entry rule is no longer valid. And by
2:48the way, I've chosen quantry model here,
2:50but as you can see, we also have other
2:52models such as Kim Kate 2, DeepS chat,
2:55cloud sonet 4, Grog four and three, and
2:59basically all the cutting edge models.
3:01Now, it gave us the strategy and it also
3:03explained it pretty well. Okay, so let's
3:05copy this, go to Jesse, and I'm going to
3:09make a new strategy. I will call it test
3:13quantry
3:15coder
3:17and paste the whole thing here. Let's go
3:19to back test
3:22coder. Set the time frame to hourly. The
3:26BTC USDT is fine. Exchange is Binance
3:28perpetual futures. And for the duration,
3:32let's select the first 6 months of 2024.
3:38The fast mode is enabled. and so is the
3:40benchmark feature. Now before I run it,
3:42let's take a look at the code. So it
3:44defined one property for the moving
3:47average of 50. It is simply returning
3:49TAS SMA which is the correct syntax in
3:52JC for using the SMA indicator. It's
3:55passing the current candles and then the
3:57period which is 50. It's doing the same
3:59thing with 100. Our entry rule is if the
4:02moving average of 50 is bigger than 100,
4:04go long. Do the opposite for short
4:07positions. And for position sizing is
4:09passing the entire balance, the current
4:12price and setting the fees of the
4:15current exchange. Then it's submitting
4:17the current buy order to open a long
4:18position using first this quantity and
4:21then the current price. So this means
4:23it's going to use a market order to
4:25enter the position. It's doing the
4:27opposite for a short position and then
4:30it's saying assuming we already have an
4:32open position and we want to update it
4:34once a new candle closes. If it's a long
4:36position and the moving average of 50 is
4:39below 100, which is the opposite of our
4:42integer rule, we want to liquidate the
4:44position. And then we do the opposite
4:46for short positions. And it is returning
4:48this as false, but it's not really
4:50necessary. We we could even delete it
4:52because we're using a market order to
4:54enter the position. But that function is
4:57only used when we are using either a
4:58limit or a suborder to open our
5:00positions. But I'm going to leave it as
5:02it is. So, let's go back and run our
5:05first back test. And it went true. So,
5:09this purple line is our portfolio's
5:12equity curve. And the orange one is BTC.
5:15It's even beating the market. That's
5:17good. It made 61% in P&L. The maxon is
5:21-7%.
5:23The V rate is 42 and the sharp is 2.12.
5:27So, this is really good and it
5:29definitely deserves a perfect 10 score.
5:32All right. Next, convert this pine
5:33script strategy into Python code. Now,
5:35we're going to need a pine script code
5:37to feed the model. And for that, I'm
5:39going to look up trading view for super
5:43trend strategy. It is this one.
5:47And if I click
5:51here, I'll be able to get the source
5:54code of it. And then we can go back and
5:57paste it here.
6:00All right. So it gave us the whole
6:01strategy. Let's copy it. All right. So
6:04it defined super trend by using the
6:08correct syntax of super trend indicator
6:10on Jesse. It is TA. Super trend pass the
6:13current candles and the period
6:15which it defined here
6:19and also the factor again here. Now it
6:23didn't really need to define it here I
6:25believe but it is fine. Now sometimes
6:27because the pine script code is doing it
6:30that way the model thinks that in Jesse
6:32it's also better to follow the same
6:34structure. So it's not necessarily wrong
6:37but usually in Jessie the syntax for
6:40strategies is significantly simpler. So
6:42you don't have to do it exactly the
6:44trading view way. Now for entry it's
6:46saying if the current trend is one and
6:50the previous trend was minus one. So it
6:52is trying to see if a crossover has just
6:56happened. Now this seems fine but it's
7:00not going to work because in just a
7:02syntax if you want to not receive just
7:04one value if you want to receive an
7:06array of values like all the previous
7:08values of that indicator you need to
7:10pass the sequential
7:14parameter
7:16and set it to true because by default it
7:18is set to false and it didn't do that so
7:22I know this is going to fail and besides
7:24that with super trend this isn't how we
7:27check if the trend is one or not because
7:30if I come here and
7:33open super trend. So basically that
7:36indicator value will give us this line
7:38here or here and if the current price is
7:40above it it means in uptrend or it
7:43should return one for the trend and if
7:45the current price is below it the trend
7:47is downward. So simply saying if its
7:51trend is one or minus one is not going
7:53to work. So overall I think this is a
7:55fail. It did do some fun stuff like it
7:58even tried to plot it on the chart with
8:00just it correct syntax. So I like how
8:03this model is thinking but this is a
8:05complete fail now because I like how it
8:09defined this correct syntax here and it
8:13also even did this. I'm going to give it
8:15one and that's a complete zero. All
8:18right next develop a mean diver strategy
8:20using the Ballinger bands and the RSI
8:22with a seven period. Let's copy this.
8:26Change the code. Paste it here. All
8:28right. So, it named it Fibonacci
8:30strategy. Define fib levels as these are
8:33the highs. The look back period is 50.
8:38And then it is passing the current highs
8:40by selecting the tree as the index. It's
8:42also fetching the recent lows.
8:45It is finding the recent swing high and
8:47low using this syntax.
8:51And the price range is highest high
8:53minus lowest low. Then it is calculating
8:55the Fibonacci retracement levels
8:59with this syntax and these magical
9:01Fibonacci numbers. Then it is passing it
9:03like this. So this is very beautiful,
9:06simple and elegant. Then it define a
9:08trend like this using two moving
9:10averages and it's saying return one if
9:13the EMA 50 is above EMA 200 otherwise
9:16minus one for a downtrend. In Schudlang
9:18first it's checking the trend. It's
9:20saying if it's minus one or if it's not
9:23one return we return false and if this
9:25passes then we want to say if the price
9:28is near or below a key Fibonacci level
9:30we want to go long again this is really
9:33beautiful like I remember previous coin
9:35models and I never like them but this
9:38one seems to be onto something. So next
9:41it's doing the opposite for short
9:43positions. And as for position sizing,
9:45it's using the current lowest low of the
9:48Fibonacci levels as the stop loss. Now
9:52let's check which one is the lowest low.
9:55Oh, okay. I see. So the lowest price.
9:57And for take profit, it's using the
10:00highest high. Now this is certainly one
10:04way to write Fibonacci strategy, but I
10:06haven't seen it. then it's risking 2% of
10:09the current available margin for each
10:11position. It's doing the opposite for a
10:13short position. Once the position is
10:15open, it is submitting the stop loss and
10:17take profit. So this is correct. But
10:20here's the thing, it didn't really need
10:22to define the takerit here. I mean, as
10:25for a stop loss, I get it. It is using
10:27it here. But this takerit variable is
10:30not being used anywhere. And in here,
10:32we're defining it even again. So this is
10:35unnecessary. And then if we already have
10:37an open position, it's saying if it's
10:39long and the current price is more than
10:43the entry price of the position plus the
10:46highest high minus the entry price of
10:49the current strategy multiplied by half.
10:52If that's the case, then update the
10:55current stop loss. So it defines some
10:57kind of trailing stop order. So this
10:59seems fine. Let's go and see if it runs.
11:02And it doesn't. So let's see where it
11:05went wrong. So, it's using this, but it
11:09didn't define it here. I mean, if it's
11:11using the 0.5 Fibonacci level in here, I
11:16believe this should also be here
11:19and here. And this should be this line.
11:24So, I think this is the correct
11:25approach. So, let's go back and give it
11:27another try. And it's not even
11:29performing that bad. Okay. So this is
11:31also a pass but I'm going to subtract at
11:35least one point because I could get it
11:38working by just one simple change and
11:40also overall the way it defined the
11:43logic of the strategy was really
11:44beautiful. Next update my code to
11:47display the values of important
11:48indicators of my strategy in back test
11:50for debugging purposes. Now for this one
11:53we're going to need a strategy and I
11:54will choose this one called teotren
11:57following from our website. I will fetch
11:58its source code. Go back and
12:03paste it here.
12:07Now let's remove these lines for
12:10the charts because we expect the model
12:14to write them.
12:17All right. So let's copy the code,
12:22paste it here. All right. So it defined
12:24this after method. It's using the
12:26addline to candle chart function. It
12:30gave it the name tema 10, tema 80 and 20
12:33and 70 for the 4 hours time frame.
12:35Everything seems correct.
12:37It define one extra line chart with the
12:41name of indicators to display the ATR
12:44value in it. And then it defined
12:46horizontal lines to our extra line which
12:49is this one. It defined the current
12:52threshold for the ADX and CMO both
12:56overbought and oversold values. But the
12:59problem is it didn't actually define the
13:01values of ADX and CMO on this chart. So
13:06it's just displaying the threshold. So
13:08this isn't exactly what I needed. So
13:10this is going to be a pass but not a
13:12perfect score. So let's go back and
13:15enable the interactive charts and give
13:18it a start. Now this error is because
13:21we're using the 4hour time frame in the
13:23strategy but we didn't define the data
13:25out but in back test Jess is smart
13:28enough to figure this out and handle the
13:30error for you. So the back test will go
13:32through but in live trading you got to
13:34be careful to add that data out
13:36otherwise your session could fail. Now
13:40it went through. Let's take a look. Now
13:42here are the values of the temo lines
13:46and this is the value of the ADX. Now,
13:49if I lower this, you can also find the
13:50thresholds for CMO and ADX. But again,
13:53this isn't exactly what I needed because
13:55the threshold alone isn't enough, right?
13:57We also want to see the values. And by
13:59the way, in the interactive charts page,
14:01you can also check out all the trades
14:03that it took and the prices and the
14:06orders. You can also click here to go
14:09and find that on the chart, which is
14:11pretty cool. So, this is a pass, but not
14:14a perfect pass. So, I'm going to give it
14:16like an eight because I believe it was
14:18an easy task and it could have easily
14:21done it. So, I'm not going to give it a
14:23nine. All right. Next. Write a trend
14:25following strategy that uses 15 minutes
14:27time frame for entry and 4 hours time
14:28frame for confirmation.
14:31All right. Let's copy it.
14:35Change the code. Paste it. And it define
14:38the long-term candles using the self get
14:40candles syntax. It passed the current
14:42exchange and symbol and passed the time
14:44frame as 4 hours. This is perfect. And I
14:47like that it first defined the long-term
14:49candles here so that it can use it
14:52later. And it is indeed using it in the
14:54long-term SMA here with the period of 50
14:58and then it is using the long-term SMA
15:00here to determine the current trend, the
15:03long-term trend really. And then in the
15:07shoot long which is the inter rules of
15:09the strategy it's saying if the short
15:11SMA is above the long SMA and if the
15:15long-term trend is simply true which
15:18would be this. So it needs to be lower
15:21than the current price then it means we
15:23are in an uptrend. All right. So it's
15:26also doing the position sizing is
15:29passing 2% as the number for risking per
15:32position.
15:34The stop is current entry which is the
15:36current price minus two times the
15:38current ATR.
15:40Then it's passing the current buy order
15:42doing the opposite for short positions.
15:45And then it's saying once the position
15:46opens if it's a long position the stop
15:49loss is going to have the quantity of
15:51the current position and the price of it
15:53is going to be the current price minus
15:55two times the current ATR. For take
15:57profit it's doing two times of that. So
16:01the risk-to-reward ratio of the strategy
16:03is going to be two.
16:05And then we're saying once every candle
16:08closes, assuming we have an open
16:10position, if it's a long position and
16:12the short EMA is below the long SMA or
16:16if it's vice versa, we want to liquidate
16:18the current position. So this seems
16:19really perfect. Let's go and run it. And
16:23it's a perfect pass. So let's give it a
16:2610.
16:32Next, here's the code for my trend
16:33following strategy in the 15 minutes
16:35time frame. The strategy is losing money
16:37because it is taking too many trades and
16:40there are too many false breakouts
16:41during the ranging markets. Please
16:43improve the strategy. So, here's the
16:45thing. Now, we want to test to see if
16:47the model has some actual trading
16:50knowledge. So, forget about coding. We
16:52want to see if it knows how to prevent a
16:55trading strategy from taking too many
16:57trades. And since it is a trend
16:59following, it needs to avoid the ranging
17:02markets. Now, for that, let's paste the
17:04previous strategy that it wrote just for
17:07us and see how it is able to improve
17:10upon it. There we go. Let's copy it and
17:13update this previous code and run it one
17:17more time. Now it's getting an error
17:20saying the source type 14 not
17:23recognized. Okay, so let's see what it
17:27changed. Okay, so here's really the
17:31problem. It's trying to define a moving
17:33average based on the values of the ATR,
17:36but this isn't the correct value cuz
17:38this parameter in Jesse is the source
17:41type of the candles. Now, if you don't
17:42know what that means, it doesn't really
17:44matter, but just know that this isn't
17:46correct. So what it should have done was
17:50to here pass the sequential
17:53as true. Now this one would have been
17:56simply the period. So 14 and then
17:59instead of passing the current candles
18:01it should have given it the ATR.
18:05So this would have worked for ATRM MA
18:09and here it could have simply checked
18:13the latest value of this with this one.
18:17So this I believe should pass. Let's
18:19give it another go. Okay, so it seems
18:23like it defined a filter but it isn't
18:27correct. So let's take a look. So it
18:30defined a couple of filters and it
18:32passed them here. But this syntax seems
18:35correct except this one. Oh, so this
18:39shouldn't have been a property. So it
18:42defined this volatility check. It also
18:45defined two more filters. So one is is
18:48the current ADX value above 25 which is
18:51brilliant. And then it's checking for
18:54choppy markets because you want to make
18:56sure the choppiness index is below 50,
18:58which is really good way to make sure
19:01there are some actual structure and not
19:03so much randomness in the current
19:05market. All right, so let's give it one
19:08more try
19:10and it did go through this time. Okay,
19:13so it got a couple of things wrong, but
19:16it did lower the number of the current
19:18trades, but it still isn't profitable
19:21actually.
19:22We aren't trading in the 15 minutes. So,
19:25let's do this.
19:29Okay, it got a little bit better.
19:32But anyways, it's not a complete pass.
19:35So, I'm going to give it a seven. Next,
19:37develop a strategy that combines the
19:39Williams percentage or MACD and
19:42Ballinger bands. Okay, I'm going to copy
19:44this whole thing and go to Jess's
19:47dashboard and paste it here. All right.
19:50So it defined the Williams R with this
19:52syntax TA. R it's passing the current
19:55candles and the period is 14. Define the
19:58MACD with this syntax which seems fine.
20:01The Ballinger bands is TA. Ballinger
20:03bands passing the current candles. Again
20:05it set the mult2 and we already know
20:08this is going to fail. So let's remove
20:10it and save it. Define the current ATR
20:13as TATR
20:15and for shoot long. Now, by the way, one
20:18thing I like about this model, it writes
20:20the code in a very clean way. So, for
20:23instance, it's wrote some really easy to
20:25read comments here. And I really
20:26appreciate this. So, it's saying if the
20:28current volumes are value is below minus
20:3180 and the maxis value is above its
20:34signal and the current price is below
20:36the current lower band value of the
20:39current Ballinger band, then we want to
20:41open a long position and do the opposite
20:43for short positions.
20:46For the sub price, it says use the
20:48current entry price minus 1.5 times the
20:51current ATR and three times the current
20:53ATR for a take profit. And for the
20:57quantity, it is risking 2% of the
20:59capital. And then it is submitting all
21:01of them here. So this seems perfect. But
21:04by the way, when I want to write a
21:06ballinger band strategy, usually I don't
21:09say if the current price is above upper
21:12band and then here you simply use a
21:15market order. Instead, I would have
21:17removed this entry rule and if these two
21:21were correct and we did come here, I
21:24would have used the value of the upper
21:26band. By the way, this one is for shoot
21:28short. This is for long. So this would
21:30have been for the short position. So
21:32instead of using the upper band here, I
21:36would have used this value.
21:38Let's copy it here for my entry price.
21:43And then all the other calculations
21:44would have worked and we would have used
21:46a limit order to open the position which
21:49would have allowed us to pay less
21:51trading fees to the exchange. But
21:53anyways, let's bring it back. And then
21:55it's saying if the MACD signal is
21:59showing a reversal or if the current
22:02price is above the middle band then
22:03liquidate the position and do the
22:05opposite for short positions. So the
22:07rest of it seems fine. It is also
22:09submitting the takeprofit and a stop
22:11loss after a position is open. Now this
22:13would have been correct if it hadn't
22:16already submitted two of them here.
22:20So I think a better syntax is to just
22:22remove them from here.
22:25So, I'm going to subtract at least one
22:27point because of this, but I believe it
22:30should work.
22:34And it does. Okay. So, it's a pass now,
22:37but I'm going to subtract one point for
22:40this mistake and another for that extra
22:43multiparameter of the Ballinger bands
22:45that we already knew doesn't work from
22:47the previous question. So, it's going to
22:48be an eight. All right. Next, write the
22:50entry functions of the strategy for me
22:52in a way that it should use two limit
22:55orders for opening the position and it
22:57should only risk 2% of capital per
23:00trade. Now, notice that I'm asking for
23:02the functions and not the entire
23:04strategy. So, what I'm hoping to see,
23:06and in fact, I am seeing it, is that it
23:09shouldn't give me the entire strategy
23:11cuz this way it would be easier for me
23:13cuz the whole idea is that sometimes
23:15you're writing the strategy and you just
23:16want one piece of the strategy and
23:19you're not sure how to write it and you
23:20ask the AI to do it for you. The first
23:22entry price is going to be 1% below the
23:25current price. The second is going to be
23:272%. The stops is going to be 2% below
23:30the first entry. So, this seems perfect.
23:32And then it is indeed risking 2%.
23:36But here's the thing. It is passing only
23:39half the current available margin. So
23:42this is going to ruin the whole
23:43calculations. So this should have been
23:45the whole thing and then it should have
23:48passed this two here. So it's a fail and
23:52the syntax for submitting the buy order
23:55is correct. So it didn't do exactly what
23:58I needed to do. So I'm going to give it
24:00an eight. Now, last but not least, write
24:02a pair trading strategy for ETH and BTC.
24:05Now, with this one, what I'm hoping to
24:07get is two different strategy classes.
24:11So, one is going to be the leading
24:12strategy which will make the decisions
24:14for opening a persing strategy and the
24:17second class is going to be the follower
24:19strategy because in JC the strategies
24:21can communicate with each other, but
24:24each one of them is going to have its
24:26own separate symbols. So we need one
24:29class to make the decisions and all the
24:31other ones but usually just one more to
24:34follow that first route. Now if this
24:37seems too complex for you make sure to
24:39check out my older video called pair
24:42reading strategy from scratch in Python
24:44where I walk you through writing a pair
24:46writing strategy in Python with Jesse.
24:49All right. So let's take a look. It
24:50defined EBTC pairing strategy.
24:54Let's scroll down and then it define the
24:57BTC follower. So let's copy this first
25:00one here.
25:03Create a new strategy and call it test
25:09three
25:11pairs one or leader.
25:15Paste it. Now let's take a look at it
25:17first. So it defined the ETH candles
25:21with this syntax
25:23which is correct but this syntax for
25:26fishing the current exchanges name
25:28isn't. So let's remove this right now.
25:35Let's do the same here.
25:39Then it converted the prices into
25:40returns using prices to returns utility
25:43function of Jesse. And it also uses
25:45zscore utility function of Jesse. And
25:48inside the before function which is
25:50called before any other method of your
25:53strategy. First is saying if the current
25:55index is zero close every position. So
25:59basically using the shared v dictionary
26:03which can be accessed from within all
26:05the trading routes that you have.
26:08It is defining this which will be the
26:10state of the current position and then
26:12it's making these decisions for pair
26:14trading which again if you don't know
26:15what these are watch my previous video
26:17about it and then it is setting the
26:21current margin per each position again
26:23using this shared VS and then to make
26:27decisions for opening the position it's
26:29simply using this it's checking what's
26:32the current state of these and it also
26:35defined the position sizing again using
26:37this ETH margin here. Now let's copy
26:40this, create a new strategy and name it
26:44test coin
26:46tree
26:48pairs follower.
26:51Paste the whole thing. Remove this part.
26:56Go back
26:57and copy this. Now notice that since we
27:01didn't have the imports above here, they
27:03were up here. I made sure to bring that
27:06from the previous strategy. So let's
27:08paste the whole thing. And this one, as
27:10you can see, is simply the follower. So
27:12it is only using these values from the
27:16shared wires. So it's not making any
27:18logical decisions. So let's go and run
27:24test three pairs leader. and the second
27:28one. And by the way, the order of these
27:32strategies in your trading routes is
27:34very important. So let's
27:37pick follower this time. Now the time
27:40frame could be anything really. So let's
27:42go with 15 minutes. Let's check the code
27:45again. I want to see which one was
27:47first.
27:49So let's go down. So ETH is first and
27:53BTC is second. So this one should be ETH
27:56and this one should be BTC. And we're
28:00going to get an error now because the
28:02fast mode isn't supported when we are
28:06trading multiple routes. So let's also
28:09disable this and give it a go. And this
28:12time it's going.
28:15All right. So we're losing money which
28:17is expected because it executed too many
28:20trades and we're paying a lot of money
28:21in trading fees. That's a typical issue
28:24with pairing strategies. Now, there are
28:26many ways to fix this, of course, to
28:28either use a bigger time frame or trade
28:30something else such as meme coins. Now,
28:33this is a big topic on its own, but the
28:35fact that it was able to write something
28:37that works is a pass. So, I'm going to
28:40give it a 10. Now, we ended with a score
28:42of 80. Now, this is a pretty good score.
28:46It's not awesome, especially because it
28:48wasn't able to convert this and it was a
28:50fail. And in this one also, it didn't do
28:53a great job. But overall, this is a
28:56pretty good model. Like even though it
28:59wasn't able to get everything working,
29:01when it did work, the code that it wrote
29:03was beautiful. And I definitely didn't
29:06expect this from Alibaba because the
29:08previous coin models, they were overfit
29:10to the benchmarks and I wasn't actually
29:13able to use them in my daily work. But
29:15this one, according to our test, is
29:17significantly better. Now guys, if you
29:19like these videos, please make sure to
29:21give them a like and post a comment. Let
29:23me know what you think. And also
29:24subscribe to the channel if you want to
29:26receive more videos like this. Thank you
29:28so much for watching. I'll see you in
29:29the next one.
29:33[Applause]
29:35[Music]

About this transcript

This page contains the full transcript of Qwen3 Coder tested: Is it good for algo trading strategies? by Algo-trading with Saleh, generated from the public captions YouTube serves with the video. The transcript has 4,932 words across 695 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.