Is a Perplexity Agent web search billed separately from model tokens?

Definition: Yes. In the Perplexity Agent API, each web_search invocation has a tool charge, while model token usage follows the selected model's rates. Search context tokens are also charged according to the amount consumed.

Key facts

• Perplexity's Web Search documentation says standard web search costs $2.50 per 1,000 invocations, or $1.00 per 1,000 when search_type is fast, plus model tokens.
• The Agent API pricing page lists web search at $0.0025 per standard invocation and $0.001 per Fast Search invocation, and states that tool costs are separate from model token costs.
• Perplexity's Projects and Billing documentation describes API usage as charged by tokens processed and search queries made.

How are a search call and token usage counted?

The Agent API runs web_search inside an Agent API request. The model decides when to call the tool based on the prompt and instructions, and each call creates one billable search invocation. Standard search uses search_type: "web"; Fast Search uses search_type: "fast" and has the lower invocation price documented above.

The search tool also has token budgets. max_tokens caps the total search context across results, and max_tokens_per_page caps the content extracted from each result page. Perplexity says the charge uses the exact number of search context tokens consumed, so the requested cap is not itself the billed quantity.

{
  "type": "web_search",
  "search_type": "fast",
  "max_tokens": 6000,
  "max_tokens_per_page": 1200
}

What appears in the Agent API response?

The Web Search documentation says that a response can contain a search_results output item before the final assistant message. Its example also shows input_tokens, output_tokens, total_tokens, and tool_calls_details.search_web.invocation in the final usage object. The invocation field records tool-call usage when it is reported, while the token fields describe token counts for the request.

"usage": {
  "input_tokens": 1200,
  "output_tokens": 300,
  "total_tokens": 1500,
  "tool_calls_details": {
    "search_web": {
      "invocation": 1
    }
  }
}

The pricing page describes Agent API model pricing as token based and says that tool costs are separate from model token costs. The billing page places both processed tokens and search queries within the API usage charged to a project.

How should a cost estimate treat the request?

Count the search invocations and the model tokens as different inputs to the estimate. For standard search, multiply the number of web_search invocations by $0.0025. For Fast Search, multiply it by $0.001. Add the selected model's input and output token charges, then account for the exact search context tokens consumed under the search tool's token budget rules.

The documentation pages opened for this entry do not give a worked example that maps search context tokens to one named field in the final usage object. The response still exposes the invocation count and token counts described above.

Related terms

• Perplexity
• Perplexity Sonar
• Agentic search
• Perplexity Research

FAQ

Is the search fee included in model token pricing?

No. The Agent API pricing page lists tool costs separately from model token costs. A request can therefore have both a search invocation charge and model token charges.

Does max_tokens mean that exact amount is billed?

No. Perplexity documents max_tokens as a cap on total search context and says the exact number of search context tokens consumed is charged.

Does one Agent API request always make one web search?

No. The model decides when to call web_search, and the response usage example reports a search invocation count. The number of invocations is therefore a request-level usage value to inspect.

Sources

1. Perplexity, "Web Search", 6 October 2026
2. Perplexity, "Pricing", 17 October 2026
3. Perplexity, "Projects & Billing", 18 September 2026

---

Check how a page reads to answer engines. Open the GEO score analyzer