How API is dealing with URLs in prompt

API
,
1

I haven’t been able to find an explicit answer to this question.

I’m interested in how is the API working with URLs. Imagine this prompt:

According to Wikipedia/Dinosaur URL, which is the tallest and heaviest dinosaur known.

Mi prompt itself has only 20 tokens according to openai tokenizer. But this prompt has a web page URL in it. The text from that web page has 56,625 tokens. But also that web page has other URLs inside.

My question is how chatgpt api is dealing with URLs. Which of the following options is the correct one?

  1. My prompt is only 20 tokens.
  2. My prompt is 20 tokens + 56,625 tokens from provided URL
  3. My prompt is 20 tokens + 56,625 tokens in the provided URL + N tokens from useful URLs inside the provided URL
  4. Other (please explain)

Thank you