YT YouTube2Text

YouTube2Text → Transcripts → GPT-5 tested: Is it good for algo trading strategies? (surprising results!)

GPT-5 tested: Is it good for algo trading strategies? (surprising results!) — Transcript

by Algo-trading with Saleh · 4,616 words · 643 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Hey guys, so OpenAI just released GPT5.
0:04Now, I've been waiting for this model
0:05for more than a year. But as always,
0:07what we care about on this channel is
0:09the performance of these models for algo
0:11trading, specifically for writing
0:13strategies in Python, debugging, and
0:15whether or not these models have some
0:17actual trading knowledge to help us
0:19improve the result of our strategies.
0:21So, in this video, we're going to test
0:22this model. If that sounds good, let's
0:24get right into it. All right. So, it's
0:26been just a few minutes since they
0:28released the model and this blog post.
0:30And according to the benchmark they are
0:32releasing here, this model is beating
0:34even the O3 model which was supposed to
0:37be really good. And on the Ader Polyglot
0:40benchmark, you can see that it is
0:42scoring 88%. That is absolutely crazy.
0:45So, if I'm not mistaken, this is even
0:47beating Claude Opus. And the price of
0:49this model is significantly cheaper than
0:51Claude. So if it is as useful as the
0:53benchmarks are claiming here, then it's
0:56going to be awesome. So let's get to
0:58testing it. Now for this video, I'm
1:00going to use GSGPT, which is the
1:02cloud-based service specifically made
1:04for algo trading in Python. So it gives
1:06us access to these models such as GPT5
1:09or any other models, but it has some
1:11instructions for algo trading
1:12specifically. Signing up on this service
1:14is completely free. But if you don't
1:16want to use this service, let's say you
1:18have some privacy concerns, or maybe you
1:20want to use the premium models, but you
1:22don't want to pay us anything, that is
1:23perfectly fine for that. Come and check
1:26out this video on my channel where I
1:28show you how you can self-host this
1:30service all by yourself. So all the
1:32instructions are open source, so you can
1:34make this all by yourself, but I'm going
1:36to use this GPT, which is just more
1:38convenient for me. Now, we're going to
1:39ask 10 questions from the model. So
1:41let's begin. Now, by the way, make sure
1:43the open AI GP5 is selected. We also
1:45added support for GP5 mini, right? Write
1:48a golden cross strategy that uses the 50
1:50and 100 moving averages. Close the
1:52position when the entry rule is no
1:53longer valid. All right. So, it did
1:56write it, but the formatting is a bit
2:00off. So, this is the first time I'm
2:02seeing this. All right. So, let's go to
2:04Jess's dashboard and create a new
2:06strategy. And I'm going to name it test
2:08GPT5.
2:11and paste the whole thing here. Now, the
2:13formatting is a bit off. I don't know
2:15why this happened. So, I'm going to fix
2:18this manually.
2:21All right. So, I believe this should
2:23work. Okay. So, let's give it a read.
2:24So, it defined a moving average with the
2:26syntax. SMA for an SMA with the period
2:31of 50. By the way, this should also be
2:33like this. It defined another one with
2:36the period of 100. It passed the current
2:38candles which is correct syntax in
2:40Jesse. It defined a property for trend
2:42return one if the moving average 50 is
2:45above 100 otherwise minus one for a
2:47downtrend. So that's perfect. The shoot
2:49function which is the in rule of the
2:51strategy is simply if the current trend
2:53is one and then in the golang which is
2:55where we do the position sizing it's
2:57first determining the quantity using the
2:59size to quantity utility function of
3:01Jesse. It is passing the entire balance
3:02and then the enterprise which is the
3:04current price. So this is correct. Then
3:06to submit the actual buy order is saying
3:07self buy equals quantity and then the
3:09current price. So this is going to
3:11execute the market order. It is now
3:13shorting which we didn't ask and it's
3:15also listening to this part where we
3:17said you know stop the position or trade
3:19if that condition is no longer valid. So
3:21in the update position which is called
3:24when we already have a open position and
3:25a new candle has closed. It is checking
3:27if the current trend is minus one or a
3:30downtrend then liquid the current
3:31position. So this is perfect. So let's
3:33go and run it. So let's change this into
3:37test GPT5 and this time frame symbol and
3:42period is fine. Let's also enable fast
3:45mode.
3:49There we go. So this is perfect. All
3:51right. Next, convert this pin script
3:52strategy into Python. All right. So for
3:54this part, I'm going to need a strategy.
3:56So on trading view, I will look up super
3:58trend strategy.
4:00And to get the source code of it, I will
4:03click here.
4:05go back and paste the whole thing.
4:09All right. So, by the way guys, uh the
4:12GPT5 couldn't get this process done. I
4:15don't know why. Maybe it's because they
4:16had just released the model and the API
4:18is heavy. I don't know. But I had to
4:20change the model from GPT5 into GPT5
4:23mini in order to for it to just go
4:25through. So, I mean, if this is going to
4:28be a pass, GPT5 should definitely be a
4:30pass, right? Because this is a smaller
4:32model. Now let's see if it works at all
4:43and it does. Okay, so this is really
4:45cool. Okay, so this is a pass. Now let's
4:48change it back into GP5 again. All
4:51right, next. Develop a mean diversion
4:53strategy using the Ballinger bands and
4:55the RSI with the seven period. All
4:57right, there we go. So let's copy
5:00the whole thing.
5:05update the code and run the back test
5:08again.
5:10There we go. So, this is a perfect pass.
5:12Now, by the way, let's take a look at
5:13the code. So, it defined the Ballinger
5:16bands with the syntax tinger bands and
5:18then it passed the current candles with
5:19a period of 20. It knows this is the
5:22default and it's also telling us about
5:23it in the comment. So, this is pretty
5:25cool. Next, it define the RSI 7 simply
5:30TA. RSI it passed the current candles
5:33and the second parameter which is the
5:35period is seven the ATR TATR the perfect
5:39syntax of JC framework then it defined
5:42the inter rules is saying if the current
5:43price is below the lower band of the
5:45Ballinger bands and the RSI 7 is below
5:4930 you should go long the opposite for
5:52short positions and as for position
5:53sizing is saying the entry is the
5:55current price the stop is the entry
5:58minus 2.5 times of the current ATR as
6:01for the quantity it's using the risk to
6:02quantity utility function is passing the
6:04current available margin 2% to risk per
6:07each trade entry price stop and then the
6:10current fees of the current exchange and
6:12then it's submitting the buy order. So
6:14it's doing the opposite for short
6:16position. So this is perfect typical
6:18strategy. Then it's saying once the
6:19position is open if it's a long position
6:21submit the stop loss and take profit for
6:23the quantity use the current positions
6:25quantity and for the price use the entry
6:28price of the current position minus 2.5
6:30times the ATR and for take profit you
6:33want to use the middle band of the
6:35Ballinger band. So this is a perfect 101
6:39mean division strategy in Python. As for
6:41update position, if it's a long position
6:43and the current price is above middle
6:45band or the RSI is above 50, we want to
6:47liquidate the position. And the opposite
6:50if it's a short position, right? So,
6:53perfect. It also defined some values for
6:56the chart so we can debug the strategy
6:59easier. So, this seems perfect. Now, it
7:02was a pass. So, let's go and give it
7:05perfect 10. Next, write a strategy that
7:07uses Fibonacci retracement levels for
7:09entry and exit points. All right, so
7:11again, it failed and I'm guessing it's
7:14because the API is under a lot of load
7:17right now. So again, I'm trying with
7:19GPT5 mini because I believe this model
7:22is also really good. And you know, if
7:25this model gets it right, then we know
7:27GP5 will definitely get it right. Now,
7:30another interesting thing is that the
7:32model is super fast. like I haven't seen
7:35anything this fast.
7:40All right. So let's paste it. Now it
7:42defined the hyperparameter. So it is
7:44assuming that we want to do
7:45optimization. That's fine. It defined
7:47the highs as the current candles and
7:50then it is selecting a certain number of
7:52them. So it defined a look back and the
7:55default value for it is 100. That makes
7:57perfect sense. And then it is select the
7:59high values of the current candles. So
8:01this is the correct syntax. and index in
8:04Jesse for low it is four and then swing
8:07high is the max of these max values
8:10swing low is the low of these I mean the
8:13min of these low values the range is
8:16swing high minus swing low if swing high
8:19is bigger than swing low otherwise zero
8:22okay so I guess this makes sense now
8:24here's the important part it defined the
8:26fib levels as range equals self range h
8:29equals self swing high l equals swing
8:32low and then the magical numbers of
8:35Fibonacci. So this is also really
8:38interesting syntax how it is simply
8:40returning a dictionary and then it's
8:42saying high minus low multiplied by
8:44these numbers. So this is very beautiful
8:48and easy to read math. Then you define
8:50two moving averages with periods of fast
8:52and slow. So we can find these values
8:55here. So 50 and 200. Then define the ATR
9:00the trend. So it's using the moving
9:03averages for the trend and then it's
9:04getting the nearest fib and in the shoot
9:07we're saying if the current trend is not
9:10one return false. So okay I didn't have
9:13to do this. I don't usually do it like
9:15this. It could just simply put it here
9:17but anyways. So this is perfectly fine.
9:19And then it's saying if the current
9:21price is below the current level plus to
9:26which is the entry tolerance and this.
9:30Okay guys, so this is really beautiful.
9:33It's already prepared for optimization
9:35and it also makes sense. So I think this
9:37is the best that I've seen so far even
9:39though this is GPT mini. So yeah, this
9:44is really good. So let's run it.
9:48and it goes through and it's perfect. I
9:51mean, it's not even performing that bad
9:53like because you know with these
9:55strategies, the raw version of them
9:58because they take so many trades, they
10:00lose a lot of money in these type of
10:03test, but this one isn't losing that
10:05much and I consider this a good result.
10:07All right, that's a perfect 10. Next,
10:09update my code to display the values of
10:11important indicators of my strategy in
10:13back test for debugging purposes. All
10:14right. So, for this one, we're going to
10:16need a strategy to give it. And I'm
10:18selecting one from our website. I'm
10:20copying the code and
10:24pasting it here. And I will also remove
10:27this part because this is actually what
10:29we expect the model to give us. Now,
10:32let's give it another try with GPT5.
10:39And it didn't work again. So let's try
10:42with GPT5 mini again.
10:46All right. So let's copy it.
10:53Update the code and let's give it a
10:55read.
10:56So it defined the after method just like
11:00we asked it to do. It put it inside a
11:03try catch block which is good practice
11:06actually. It is using addline to candle
11:09charts method of Jesse. It's giving it
11:12the name of the TMA indicator once for
11:15TMA 10 and then TMA 80. So this is
11:17great. It defined
11:19two more times for TMA in the forest
11:22time frame. Then it is defining extra
11:26line charts for ADX and CMO and it's
11:30putting both of them inside the same
11:32chart. So this actually makes perfect
11:34sense because both of them have values
11:36from 0 to 100 and then it defined the
11:39threshold lines of these indicators. So
11:42this is very beautiful guys. It's also
11:44doing some logging here. Now I didn't
11:47ask it to do that but maybe because we
11:49mentioned debugging it also added this.
11:51So this is pretty beautiful. So let's
11:53give it tries to see if it runs well.
11:56Now I should have enabled the generate
11:59interactive chart. So let's give it one
12:01more go
12:04now. There we go. So we can see the temo
12:07lines.
12:09They are beautiful and they make sense.
12:12And here we can see the values for ADX
12:17and also the CMO. Now this green isn't
12:22really visible on the dark mode. Now we
12:24could also specifically ask for a
12:26certain color when we define it with
12:28that function. Otherwise, Jesse will
12:30just randomly assign a color to it. So,
12:33don't worry about this much. But this
12:35seems perfect. By the way, in my
12:36previous test, I've seen many models not
12:39being able to do a great job with this
12:41question. So, the fact that GPT5 mini
12:44model is doing a perfect job is really
12:47awesome. All right. Next, write a trend
12:49following strategy that uses 15 minutes
12:51time frame for entry and 4 hours time
12:53frame for confirmation. I believe GPT5
12:57mini should also nail this one. So, I
12:59don't think we even need the bigger
13:00model. So, in fact, this model seems to
13:03be so good that I'm going to set this as
13:06a default model for new users because
13:08not only is cheap, it's also pretty
13:10great and really fast.
13:13All right, so let's copy the whole
13:15thing,
13:21update our code. And by the way, I've
13:23seen this before. Sometimes it adds an
13:26extra py here which breaks the
13:29execution. So let's remove that one. All
13:31right. So it defined the big candles as
13:33a separate property using the get
13:36candles method of Jesse passing the
13:38current exchange symbol and 4 hours time
13:40frame which is what we ask for. It
13:42defined it like this so that later it
13:44can use it twice. So that makes perfect
13:46sense. It already prepared the strategy
13:48for optimization as well. And then it
13:50defined an ADX filter saying that the
13:54current ADX must be above the threshold.
13:57Now let's see what's the value for it.
14:00So by default it is 25. Okay. And next
14:04it defined the in rule of the strategy
14:07as the big EMA short must be above big
14:10EMA long. So we're waiting for a
14:13crossover basically. And then the EMA
14:15short must also be above EMA long and
14:18ADX must be okay. So this okay word that
14:21added at the end of it kind of clarifies
14:24that this is a filter and I actually
14:25like that. So everything else seems
14:28fine. Let's go and give it a test. And
14:32it runs perfectly. So again this is a
14:34perfect 10. Now no model so far had
14:38gotten a perfect 10 score on the first
14:40six questions. So, so far this is
14:43beating everything. All right. Next,
14:44here's the quote for my trend following
14:46strategy in the 15 minutes time frame.
14:47The strategy is losing money because it
14:49is taking too many trades and there are
14:52too many false breakouts during the
14:53ranging markets. Please improve the
14:55strategy. So, we're talking about this
14:56same strategy that he just wrote for us.
14:58You see, it's losing money. Now, by the
15:00way, I didn't execute it on the 15
15:02minutes. So, let's do this. Let's also
15:06disable the charts so that it would go a
15:08little bit faster.
15:11All right. So, you see it's losing a lot
15:12of money. Why? Because it's taking too
15:14many trades. So, 561,
15:17which is too much because we're paying a
15:19lot of money in trading fees. So, if we
15:21add this fees into here, we will end up
15:24profitable, right? But we are going to
15:26have trading fees in actual trading. So,
15:28we need some filters to only take the
15:30good trades to prevent false breakouts.
15:33Right? So with this question, what I
15:35want to test is to see if the model has
15:37some actual trading knowledge to know
15:39about this to, you know, what to do in
15:41these type of situations. So let's go
15:43back and copy this code. And I'm going
15:46to paste it here.
15:50All right, there we go. So let's copy
15:51this. Go back, paste it again, and
15:54remove this extra py.
15:57And let's take a look.
15:59So this is a should our entry rule. So
16:02it defined a couple of filters. So one
16:05is recent trend gap. Okay. So let's see
16:08what this is. All right. So it's
16:10checking if the current index minus the
16:12last 30 index is bigger than a certain
16:15number. So it's basically checking to
16:16see some time has passed since our last
16:19trade. So this makes perfect sense
16:20because you know sometimes especially in
16:22trend following strategies we try to
16:24escal a certain trend and then and we do
16:27this a couple of times but sometimes you
16:29know maybe after two or three times that
16:32trend is about to finish right so it's a
16:34good idea to say that okay if you took
16:37already two trades very fast then don't
16:40take a trade for at least 10 more
16:42candles so I've seen this in practice so
16:45this is great it also define another
16:47filter called breakout confirmed med
16:49long. So it's checking to see if the ATR
16:52multiplied by a certain number plus the
16:54current donin's upper band is below the
16:58price. So so to put it simply we're
17:00using the donin channel and we want to
17:02make sure that the current price has
17:04broken above that. Next it did the
17:06opposite for short trades and then we
17:10also have one volatility
17:13filter. So let's see how is this one. So
17:16it defined the BBW which is Ballinger
17:18bands weight and it and it needs its
17:20value to be above a certain value. So
17:22this is also really good because when we
17:23are in a range market the value for the
17:26Ballinger bands width is tends to be
17:29low. So this is also a great one. And
17:32the next one is ADX which is simply if
17:35the ADX is bigger than its threshold.
17:37All right. So it defined two ADX for
17:40both the current trading time frame and
17:41the bigger time frame. So this is really
17:44great. All right. So, this is perfect.
17:45Even if it doesn't run, I'm going to
17:47give it a good score, but I believe it
17:50should. Okay, I think I forgot to save
17:52the code. All right, so let's save it
17:55this time. And go back and run it again.
18:01There we go. So, now it only took nine
18:03trades instead of that many. And it's
18:05not losing as much. The draw down is
18:08also pretty good. Okay, so it's not
18:10profitable, but it didn't take as many
18:12trades. And I really like the filters
18:13that you define. Okay, so this is
18:15another perfect 10. And guys, this is
18:17GPT5 mini, not the bigger model. All
18:19right, next. Define a strategy that
18:21combines the Williams percentage or MACD
18:24and Ballinger bands.
18:26Now, considering what it did with the
18:28previous questions, I think this is a
18:30very easy one for this model. And you
18:32see how it just appeared. So, it is
18:35really fast. So, apparently, it's a
18:37thinking model. So it does some
18:39reasoning and then it outputs the result
18:41almost instantly. So that's pretty
18:43helpful. Okay. So let's copy it.
18:50Okay. So let's paste it. It defined the
18:52Williams R with this syntax which is
18:55correct. The MAC is correct. Ballinger
18:57bands the ATR and for the entry rules
19:00it's getting the upper middle and lower
19:02band from the current Ballinger bands.
19:05So this is also correct. And then it's
19:06checking if the current closing price is
19:09below the lower band and the Williams R
19:11is is below a certain threshold and the
19:14MAC H is above zero. Okay, so this seems
19:18perfect. Let's go back and give it a
19:22try.
19:24And there we go. Okay, so this is again
19:27a perfect 10. Now it's still it is
19:29beating all the other models so far.
19:31Right. Next, write the entry functions
19:33of the strategy for me in a way that it
19:35should use two limit orders for opening
19:37the position and it should only risk 2%
19:39of capital per trade. So, I want to see
19:41whether or not it will just give me the
19:43functions, not the whole strategy
19:44because sometimes you need one specific
19:46function for your strategy. You want to
19:48copy and paste it and use it very
19:50quickly. So, a model which prints out
19:53extra stuff isn't really useful in these
19:55type of cases. And also I want to see if
19:57it is able to use the correct syntax for
20:00opening the position using multiple
20:02orders. So not just one market order
20:04like all the previous questions. All
20:05right. So it did give it to us but it's
20:08not pretty. The formatting is incorrect.
20:10So let's ask it to fix it.
20:19All right. So let's see if it will fix
20:21it. Okay. It didn't. So, this isn't
20:23great because it's going to be a bit
20:25hard for me. Now, I don't know why this
20:27is happening, but here's the thing. This
20:29might be just on our website. So, I'm
20:31not sure if I should subtract points
20:33from the model because of this. All
20:35right, but just in case, let's just copy
20:37this whole thing. Go to GPT and I will
20:40simply ask it to fix the formatting,
20:44please. And I didn't even write it
20:47correctly. And let's see what it will
20:49give me. Okay, so this is an easy task
20:51for JBT and you know hard for humans to
20:54manually fix something.
20:56Okay, so it's defining the current ATR.
21:01It defined one entry and two and it's
21:04using the ATR for setting the price and
21:07then it defined the stop price. Okay, so
21:09this is all good. It is using risk to
21:11quantity utility function of Jesse. So
21:13this is the correct syntax for risking
21:152% per each trade on with Jesse. So it's
21:19it's even defining the precision which
21:21usually the other models didn't. They
21:23just went with the default. Now the
21:24default value is fine but this is
21:27interesting. Then it defined the two
21:30quantities and it's also ensuring the
21:34quantities are bigger than zero. So this
21:37is also good practice but you know not
21:39necessarily at all. And at the end it is
21:41submitting both orders using this
21:43syntax. So instead of giving it just one
21:45tupil of values, it defined one list and
21:48inside it it defined two tupils. So
21:51first the quantity and then the price.
21:52So this is perfect and it did actually
21:54just give us the functions. So I'm going
21:56to give it a pass.
22:00All right. So last question. Write a
22:01press rating strategy for ETH and BT.
22:03Now because this is a tough one, I want
22:05to give GPT5 another try. Okay. So it
22:08fell again. Now again guys, by the time
22:11you're watching this video, this issue
22:13has probably been fixed. So don't worry
22:16about it much. But I'm going to try it
22:18with GPT5 Mini because so far it was
22:20able to solve all the questions and it
22:23got a perfect 10 for all of them. So
22:25maybe it will get this one right too.
22:27Now before I continue, I got to say this
22:29is a tough one because it's a pair
22:31threading strategy and the model is
22:33supposed to define two classes. The
22:35first class is the one that has all the
22:37trading logic in it. And the second one
22:40is simply follower class because it
22:42needs to execute the decisions that the
22:45master strategy class has made for it.
22:47And the code of it is also a bit
22:49complex. So if you're not familiar with
22:52pair trading, go ahead and watch this
22:54previous video of mine called praing
22:56strategy from scratch in Python where I
22:58go through the basic math and coding
23:00behind this strategy. But anyways, let's
23:03copy and see if this will work.
23:09Okay, so I'm going to define a new
23:11strategy and call it test GPT
23:15pair one
23:20and this is the follower. So we don't
23:23need this one. Now this seems perfect.
23:26It define hyperparameters. Okay, so this
23:28might be an issue
23:31because you see when we use multiple
23:33trading routes, we don't do optimization
23:35on it. So I don't know if this will
23:37work, but it's not really the model's
23:39fault, I guess. Let's make sure this is
23:41saved. And then define the second one.
23:44Test GPT pair 2. Paste the whole thing.
23:51And this time I only want this second
23:54class. So let's res it. And this also
23:59seems fine to me. Okay. So let's run it.
24:05The first one is BTC. The second one is
24:08ETH. And let's change this to be hourly.
24:14Test
24:17GPT pair one
24:20and test GPT pair two.
24:24All right. And we also need to disable
24:26the fast mode because we're trading
24:28multiple trading routes.
24:35There we go. So, yeah, it works. Now,
24:38it's losing a lot of money. It doesn't
24:39matter because it took so many trades.
24:41We're paying a lot in trading fees,
24:42which is also a typical issue with the
24:44pair trading strategy. And that wasn't
24:46really what I was looking for. So the
24:48fact that it wrote it fine and it works
24:52means it will get a perfect 10. And look
24:54at this. It got a perfect 100 score. And
24:58let's not forget for most questions I
25:00use GPT5 mini. Not even the GP5 itself
25:04which was supposed to be the model we
25:06were going to test in this video. Right?
25:08So even the smaller model is nailing all
25:11the questions. And guys to give you a
25:13comparison the Grog for model when I ran
25:16the same test it scored 85. Now the
25:20second question wasn't running. So even
25:22if we gave it the perfect 10 here still
25:25the ending result would have been 95.
25:27And the cloth for set which to this day
25:30was my go-to coding model wasn't able to
25:33answer all the questions. Now back then
25:36we weren't using a numberbased system
25:38but still three of the questions were
25:40fail. So again, the GPT5 mini model
25:44according to my tests is beating not
25:47only Grock 4, the big model, and it's
25:50also beating cloth for Sonnet. This is
25:53crazy because GPT5 mini is extremely
25:55cheap and not only that right now I'm
25:58wondering if this model is so good, how
26:00good is GPT5 itself. So yes, according
26:03to these tests, we didn't even get to
26:04need a bigger model, but one day we
26:07will, right? So, let's say you are
26:08really stuck and GPT5 mini isn't able to
26:11help you. In that day, GPT5 is going to
26:14be even better. So, just wow. Like, I
26:16don't know what to say. Great job,
26:18OpenAI. You guys really surprised us,
26:20and I'm glad the wait was worth it. So,
26:22anyways, if you guys want me to make
26:24more videos like this, please like the
26:25video, post a comment, let me know your
26:27opinion, and subscribe to the channel.
26:29It will help me out a lot. Thank you so
26:31much for watching. I'll see you in the
26:33next one.
26:43[Music]
26:44[Applause]
26:47[Music]

About this transcript

This page contains the full transcript of GPT-5 tested: Is it good for algo trading strategies? (surprising results!) by Algo-trading with Saleh, generated from the public captions YouTube serves with the video. The transcript has 4,616 words across 643 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.