Perplexity has changed the default retrieval mode used by the Agent API's web_search tool for the low, medium and high presets. According to the October 2026 API changelog, these presets now use search_type: "fast", matching the existing fast preset. The xhigh preset continues to use standard web search.
The practical impact is cost and latency: Perplexity says the affected web-search calls drop from $2.50 to $1.00 per 1,000 invocations, with approximately 800 milliseconds less latency. Developers who need the standard retrieval path can explicitly set search_type: "web".
Which presets changed?
| Agent API preset | Default web search mode |
|---|---|
low | fast |
medium | fast |
high | fast |
fast | fast (already supported) |
xhigh | Standard web search |
This is a change to the presets' default retrieval behaviour. An integration that relies on a preset without explicitly setting search_type may therefore take a different search path even if its own code has not changed.
Cost and latency impact
Perplexity lists the new Fast Search rate for these Agent API web_search invocations as $1 per 1,000 calls, down from $2.50 per 1,000 for standard web search. That is a 60% reduction in the per-invocation search-tool cost. The company also reports roughly 800 milliseconds less latency.
These figures refer to the web-search tool invocation, not the total cost or end-to-end latency of an entire agent run. Model tokens, additional tools, retries and the number of searches performed can still contribute to total usage and response time.
How to retain standard web search
Developers who prefer the standard retrieval path can explicitly set search_type: "web" in the web_search tool configuration. Setting this parameter avoids relying on the preset's default for the retrieval mode.
Before changing production integrations, review the current API schema and test representative queries. Compare returned sources, freshness, coverage and answer quality across both modes; a lower price and lower latency do not by themselves establish that the results are interchangeable for every use case.
Why this matters for retrieval quality
Search configuration is part of an agent's behaviour. A default change can affect which retrieval path is used, the evidence available to the model and the time spent waiting for search results. For research-heavy workflows, teams should verify whether Fast Search returns the level of coverage and source diversity their task requires.
The changelog establishes the preset change, price and approximate latency reduction. It does not claim that the change improves or degrades result quality across all queries. Quality, source freshness and citation differences should be measured in the specific workflow rather than inferred from the preset name.
Recommended checks for API teams
- Inspect preset usage: identify integrations using
low,mediumorhighwithout an explicit search type. - Make retrieval mode intentional: set
search_type: "fast"orsearch_type: "web"when consistent behaviour matters. - Run paired tests: compare the two modes on representative queries, including time-sensitive and research-intensive tasks.
- Measure the right costs: separate search-tool charges from model-token and other tool costs.
- Audit evidence quality: check source relevance, freshness, diversity and citation support before assuming the cheaper mode is sufficient.
- Document the change: record the mode used so later differences in output can be interpreted correctly.
Frequently asked questions
Which Perplexity presets now use Fast Search?
The low, medium and high presets now use search_type: "fast" for Agent API web_search. The fast preset already used this mode.
Does xhigh also switch to Fast Search?
No. The October 2026 changelog says xhigh remains on standard web search.
How much does the change save?
The reported web-search invocation cost falls from $2.50 to $1.00 per 1,000 calls, with approximately 800 milliseconds less latency.
Can developers still use standard web search?
Yes. Explicitly set search_type: "web" to request the standard retrieval path.
Does Fast Search guarantee the same sources or quality?
The changelog does not establish identical results or quality across all tasks. Teams should compare source relevance, freshness and coverage for their own workloads.
The takeaway
Perplexity's Agent API now routes the low, medium and high presets through Fast Search by default, reducing the listed web-search cost to $1 per 1,000 invocations and latency by about 800 milliseconds. xhigh remains on standard search, and developers can preserve that retrieval mode explicitly with search_type: "web". The change makes it worth auditing preset-dependent integrations and validating source quality before assuming the faster path fits every task.
Primary source
Perplexity API Changelog (October 2026).