Issue you'd like to raise.

costs shown on langsmith for perplexity API calls are inaccurate (even upto an order of magnitude). refer to their docs for how to calculate costs: https://docs.perplexity.ai/guides/models/sonar-reasoning-pro#detailed-pricing-breakdown-for-sonar-reasoning-pro

The problem is that pplx cost depends not only on input and output tokens but also on search queries and

Suggestion:

either add support for calculating this within langsmith platform or let users update the true cost while tracing an LLM call.