Best llm model or api for text translation (output token limit?)

Hi everyone, I am currently learning about youtube and website downloading.

In this context I also like to have the possibility to automatically translate any foreign language audio and text. Having the possibility to process 50.000 token in one go (dont feel like chunking), translate whole books, podcast captions, ...

Sure, I can use "OpenAI Audio Translate" instead of "OpenAI Audio Transcribe", but I also like to have a possibility to just translate text.

I have tried:

1) DeepL API: Pretty nice, much faster than OpenAI, but once you get over the 500.000 free per month tokens also very expensive

2) OpenAI GPT: Just using ChatCompletions and an appropriate prompt. My problem: The output token limit for most models is 4k. The models with 8k or 32k are super expensive.

I have not tried google translate yet.

Do you know any good models or api that are fast have good quality, fast and a large output token limit?

P.S. gpt4o has 128k context window, but only 4k output token limit, most models are like this.

Does anyone have a good idea?

12 comments