0:00 To use cloud code, you could either sign up to the pro plan, a max 5x plan, or a max 20x plan. But these aren't the only ways. You can also use their APIs. But which is cheaper? This will be a bit of a nerdy explanation. So if you just want the direct answer, best thing you can do is just get propan and then if you ever run out of tokens, you then use the API. But if you're like me and you like to have a calculation for each token that will be spent, then follow along. I'll show you exactly the results I got. So, first of all, I don't know if it's just me or if more people got confused with this, but this pro plan is the Claude pro plan. It isn't just a Claude code plan. And I thought this just because we're inside of the Claude code page right here and suddenly we have all these options along with Claude code is
0:45 included in your pro plan. So, initially I thought, wait, isn't this the Claude code plan? But no, this is the actual like claude plan that you have access and all the tokens that were to be spent in here will be instead consumed in your cloud code. So the information on anthropic gives us about the pro usage is that it has 5x the usage compared to the free service. And the max 5x plan has five times the amount of tokens that we'd get in the pro plan. To estimate how many tokens I spent per month, I'm using the cursor dashboard. This used to be much better. It used to have I believe it used to have token counts but now you just get the amount of lines that the agent edited. And to get an estimation of how many tokens that is, I just multiplied that by 75, which is
1:32 what cursor told me is the average number of tokens that each line of code has. This gives us an estimate of 11 million tokens usage per month. So that's an average of 375,000 tokens per day. And just now I realized that I divided by 30 days and then I used 22 days of working days. But this is just for my specific case where I'm using 11 million tokens of estimation. But I'll also make a comparison of how many tokens you could have used per month for each specific plan. If you're wondering what are these two 5our cycles per day, this is the amount of time that Claude takes to refresh your tokens. So if I send a message to Claude, it'll start counting from now until 5 hours from this exact moment. During this time, you'll have an x amount of tokens to spend. And the whole problem with this calculation is that we don't know exactly how much tokens we can spend in
2:18 this range of time. So the rest here is just registering how much we'll spend per API. And we're considering the cloudset 4.0. We'll also consider prompt caching as well as batch processing. So let's start off with the token limits per 5hour window. The first website that I found talking about this estimated that each fivehour window allows us to use a total amount of 1,600,000 tokens. And I know this is Claude 3.7 and not the 4.0, but the pricing is exactly the same. The API price is the same. And therefore, I presume that the amount of tokens anthropic gives us is basically the same. And when I read this, I thought, no way this is the right price. Like a,600,000 tokens is a lot. But then he mentioned that he was using a Claude usage tracker which is this Chrome extension right here that
3:03 allows us to see how much tokens were spent per chat. So because of that I followed along with the calculation. So the pro plan would have those 1,600,000 tokens. The max plan would have 8 million tokens. And since the beginning I wanted to use a two window of these 5 hours because when you start to use it right here you'll have an X amount of tokens. Uh after you go through this you'll get into the second part. So these are two windows of token refresh. The first one you start off with those x amount of tokens, then you start off again right here with a refreshed amount of tokens. So it's not like if you have to wait like 10 hours to have more tokens, you just have to wait five. So yeah, you got the point. That's why I multiply everything by two. Considering that every day we'd use two cycles of the token refresh and then considering
3:51 working days, we'd have 70,400,000 tokens for the pro plan, which then gets us to the comparison with the API cost at $718, which is completely unrealistic. So then I proceeded to install the extension and test it for myself and discovered that it's not tracking the amount of tokens correctly. So that is definitely not the amount of tokens that Cloud gives us per 5hour window. That took me to this Reddit post which estimates 27,000 to 32,400 tokens per cycle which is much more realistic. That would end up giving us a,320 tokens per month in the pro plan which is around $1346 for the API cost. As for the max plan, that would be 6,600,000
4:36 tokens and a estimated 67.32. So, at this point, I had this table, which basically tells us that using the API is much better than signing up the plan. But still, this number was based off a Reddit thread, so I felt a bit insecure making a video just based off of this number. So, I'm not really proud of myself for this, but I went over to Claude and drained all my tokens just to count how many tokens they give us on the free plan. And after I had that number, I just had to multiply it by five. This is the website that I used to calculate the amount of tokens I spent uh specifically for the cloud sonet 4 which estimated around 20,000 tokens 7,278 words. And this is only for the output since as you can see the inputs are really small. So I didn't even consider those. Uh I basically asked it to create
5:22 like long things. So please create an entire website just using HTML. It should be an admin dashboard using shaden components. It failed to do so because since it was too big, it didn't finish creating the HTML. Then I asked it to explain things in full detail. I mean, this isn't the best way to test this, but it is an estimation. And since the free plan has around 20,000 tokens, that means that the pro plan is 5x that number, which is around 100,000 tokens. And if that is true, considering the two cycles each day and the 22 days, we have around 4,400,000 tokens per month. And with this estimation, I also consider that the a number of input tokens would be 40%, the output tokens 60%, which for coding, I'd
6:08 say that we have much more output tokens than input tokens maybe it would be a parola law of 20% with 80%. But the estimation I got for this is $44.88, which honestly I believe is closer to the real number just because I had used cloud code before as well as the API and I feel like the API wastes much more. And finally down here in the comparison with the amount of tokens that I actually spent per month with which is around 11 million. It would be insufficient to use the free plan for sure. Uh the pro plan is insufficient as well as the pro plan gives us around 4,400,000 tokens per month. But the max plan would be sufficient and actually like double the amount. So, like I said in the beginning of the video, going for the pro plan and then whenever the limit is
6:54 reached, you go for the API should be the better approach, unless you're doing pretty small things that wouldn't reach the $20 mark with the API. So, yeah, that is it for today. Thanks for watching. If you have your own calculation for this, please let us know down in the comment section and I'll see you in the next video. So that
