Low, medium, high, and xhigh presets use Fast Search
By Steven Van ·
Agent API web_search calls under the low to xhigh presets now cost $1.00 per 1,000, down from $2.50, and run about 800 ms faster.
Perplexity's Agent API now runs web_search on Fast Search for the low, medium, high and xhigh presets, matching the fast preset. These presets now set search_type: "fast" by default.
Price and speed
Under these presets, web_search calls drop from $2.50 to $1.00 per 1,000 invocations, plus model tokens. Search is about 800 ms faster. Models, prompts, tool budgets and step limits are unchanged.
Keeping standard web search
To use standard web search under any preset, set search_type: "web" on the web_search tool in the request. If search_type is omitted, the tool inherits the preset's search type when one is configured, and otherwise uses "web". The docs suggest fast for day-to-day agent loops and high-volume tool calls, and web for rare, difficult or ambiguous questions.
The setting is separate from the Agent API fast preset. It can be combined with token budgets, max_results and filters. The Python library (0.43.5) accepts search_type inside the tool object. In the TypeScript SDK (0.38.5) the tool type does not yet include the field, so the docs use an as any cast. Details are in the Agent API Web Search documentation for Perplexity.