So I want to confirm if my cost computation is right because the difference seems to be much greater from what I’ve read online.
I want to compare DeepL and ChatGPT for machine translaton and tried to estimate costs.
For DeepL, I performed the translation task using 10 * 2k sentence into one language which got cost about 80 Euros.
I have to do this task 11 times, so it would cost me 880 Euros. For DeepL, I just compute the costs using character counts. But I know for sure that it took 80 Euros when I translated 10 * 2k sentences, I saw the usage cost on my dashboard, so I have some confidence that cost estimation is close to the real cost I will have in the end.
For ChatGPT, since the cost is done with tokens, I used the text 10 * 2k sentences and tokenized them with tiktoken. This yielded me about: 1,260,000 tokens. The cost for GPT4o Input is 2.5 USD per 1M tokens and the cost for output is 10 USD for 1M tokens.
So I assume, if I want to estimate the cost for translating 10 * 2k sentences into one language, I get:
(1260000 / 1M )* 2.5 + (1260000/ 1M) * 10 = 15.75 USD
And if I do this 11 times, I get 15.75 * 11 = 173.25 USD.
Which is about 5 times cheaper than DeepL (Omitting Euro / USD conversion)
Did I make a mistake in the cost computation or is this correct?