You have to be retarded lol. You just said before that everything is written on a website, yet now, since I just proved it isn't, you are saying that every single customer is supposed to ask LOL. Its like selling a giftcard, but only after the sale you get told, "well buddy you should have known it only works in one specific location, and only for this one specific item, you didn't know that, well that's your problem, I don't gotta tell you shit?" what a stupid take. You are obviously trying to mislead people, why not just make a token system that is similar to Claudes one? Why make it be inflated by 350x-700x, and never state it anywhere? Right... because you want people to be mislead and think they are getting more than they actually are.
And again, you still didn't tell me why it is double billing, you just keep on dodging that, but sure we can wait for staff.
Now that youre not so rude I can answer in details: We calculate input and output differently as I said.
We register as a startup on a certain api provider which I can't disclose! The explanation is: It does not return the exact token count—only the percentage of the context window (1M) used. The minimum is 2% (~20k tokens), even if you send just a single word. Consequently, small requests are always overestimated. The direct Anthropic API counts tokens precisely. In practice: for actual context up to ~40–50k, This specific provider inflates the figures, but accuracy improves beyond that point. If you want to verify this, calculate the tokens locally using `tiktoken` and compare the result with the dashboard. This is precisely why the new CLI service (sk-ant-cap01-...) is more transparent: it uses direct Anthropic access with precise accounting.
Service is being provided exactly as is and everyone is happy for the last month, only had to deal with you two that have don't know the service and complain expecting something unrealistic.
you said 98% cheaper which I never mentioned, said one prompt would eat all tokens but didn't even use them, he used 100k tokens I have the key and can share to the dashboard with the times he used.
Stop buying third-party API services if you don't know how it works, go with anthropic official api and use your 4$ with opus 4.8 1m context which will lasts one message. The calculation is clear you dont
Now that youre not so rude I can answer in details: We calculate input and output differently as I said.
We register as a startup on a certain api provider which I can't disclose! The explanation is: It does not return the exact token count—only the percentage of the context window (1M) used. The minimum is 2% (~20k tokens), even if you send just a single word. Consequently, small requests are always overestimated. The direct Anthropic API counts tokens precisely. In practice: for actual context up to ~40–50k, This specific provider inflates the figures, but accuracy improves beyond that point. If you want to verify this, calculate the tokens locally using `tiktoken` and compare the result with the dashboard. This is precisely why the new CLI service (sk-ant-cap01-...) is more transparent: it uses direct Anthropic access with precise accounting.
Service is being provided exactly as is and everyone is happy for the last month, only had to deal with you two that have don't know the service and complain expecting something unrealistic.
you said 98% cheaper which I never mentioned, said one prompt would eat all tokens but didn't even use them, he used 100k tokens I have the key and can share to the dashboard with the times he used.
Stop buying third-party API services if you don't know how it works, go with anthropic official api and use your 4$ with opus 4.8 1m context which will lasts one message. The calculation is clear you dont
Where does it say that on your website? You said everything is on your website in your last reply, and to read it, which I did before purchasing, where is all this stated? Also you still didn't answer why it is double billing every single prompt.
@Liars You may want to look into this API token scam. This is not the first case, and it likely won't be the last. It's a common honeypot for scammers because it can be set up in minutes and, if done properly, can look completely legitimate to potential victims.
So I tested it out as seller wanted and used exactly one prompt. Here is how it went. The prompt involved creating a skeleton for a website that includes a database and login functionality.
Claude ran for exactly 12 minutes and 16 seconds and has eaten around ~50k tokens (original Claude tokens) <- https://ibb.co/VcTcYHw5
Unfortunately, I didn't get the screenshot exactly when it ended since it disappeared, but I got one at ~6 minutes showing 25k tokens. <- https://ibb.co/4RBz0wyR
My input was barely any tokens, since it was just a single message with no history, so let's calculate!
1M tokens output on Opus 4.8 is $25, which would make ~50k tokens be around $1.25 + whatever the input was, but that was maybe a few pennies, if even that.
Now about the guy that states to provide a 93% discount on the Claude API, it was 4,548,736 tokens, which is $3.93, since I paid $4.32 for 5M tokens. <- https://ibb.co/VYTtFmRN
So where is the discount? How is it 3 TIMES MORE EXPENSIVELMAO!!! I thought the more Claude works on it, the more caches it or whatever, no? Only small prompts would be losing the discount, right? That's what you said?
Let's see the "requests" it was sending. <- 1. https://ibb.co/YBQ0kDRx , 2. https://ibb.co/bhPp6NQ
Every minute it literally took upwards of 80K tokens of input out of me. Even though I wasn't prompting anything, Claude was just working.
Also the output that we can see on those two screenshots is the correct one that Claude actually spends on output, so why is there a literal "input fee" every minute that keeps on increasing?
It started at 40k each minute until it literally got to 80k, taking in total around ~4.5m tokens just to finish one task, lol. (What I mean about INPUT, is in the two screenshots, you can see the tokens it eats on INPUT, which is users prompts, and you can see the time next to it, its unrealistic for me to spend so many INPUT tokens every minute, sometimes few seconds)
So, @MrPoisonLab , how is it possible that buying tokens through Claude officially would be over 75% cheaper than buying them from you? I thought it was 93% off, right?
after this threads why u add this steps now : // Common questions Got questions? before this i check website 10 times and i didnt see any questions like this
Again this @VelvetV7 coming to my telegram dm to try mock me!
The dispute is right here not on telegram, I blocked him.
Looks like @VelveV7 liked the service, he's been using his tokens now...
Again this @VelvetV7 coming to my telegram dm to try mock me!
The dispute is right here not on telegram, I blocked him.
Looks like @VelveV7 liked the service, he's been using his tokens now...
Do you even read our replies LMAOOOOO? I liked your service?
Your API service billed me 3x more than official Claude API would.
You was literally constantly telling me, that I didn't even test it properly, well I did, with one prompt, and it ate my whole balance, as I predicted before.
![[Image: b.png]](https://i.ibb.co/VY4h7YhN/b.png)
![[Image: y1SoGfC.gif]](https://i.imgur.com/y1SoGfC.gif)
![[Image: kek.png]](https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Fstatic.cracked.st%2Fimages%2Fsmilies%2Fkek.png)