API Pricing - how to tell how many searches a given request incurred

[ssalbiz-relay]

Based on the pricing documentation, perplexity charges $5/1000 search queries incurred by a the model.

However, I don't see a way to determine based on the model response how many searches were done under the hood - just the token counts are provided, and its implied that in some cases the model will perform multiple searches to produce a result.

Is that something that can be inferred (e.g. by counting citations maybe?) or is it simply not exposed yet please?

[AidanShipperley]

It's in the usage information in the response object. If you're using Python for instance, you can get it like so:

from openai import OpenAI
import os
client = OpenAI(api_key=os.environ.get("PPLX_API_KEY"), base_url="https://api.perplexity.ai")

response = client.chat.completions.create(
    model="sonar-pro",
    messages=[
        {
            "role": "user",
            "content": "What is the most recent version of Python?"
        }
    ]
)

print(f"Prompt tokens: {response.usage.prompt_tokens}")
print(f"Completion tokens: {response.usage.completion_tokens}")
print(f"Citation tokens: {response.usage.citation_tokens}")
print(f"Number of search queries: {response.usage.num_search_queries}")

This gave me:

Prompt tokens: 9
Completion tokens: 280
Citation tokens: 3353
Number of search queries: 1

[drodrig]

Hmm, my results omit the number of searches.

 'usage': {'completion_tokens': 478,
           'prompt_tokens': 3260,
           'total_tokens': 3738}}

I found this info searching:

The number of searches Perplexity performs per query depends on the specific API model used:

Sonar Reasoning and Sonar Models: Each request performs 1 search.

Sonar Pro and Sonar Reasoning Pro Models: Each request can perform multiple searches to provide more comprehensive information. For example, a request that performs 3 searches will cost $0.015 in this step.

EDIT: I do sometimes see the search number, but not always:

 'usage': {'citation_tokens': 4248,
           'completion_tokens': 279,
           'num_search_queries': 2,
           'prompt_tokens': 4686,
           'total_tokens': 4965}}

Not sure why.