Perplexity Fast Search API: Speed, Pricing And Setup (2026)

Set search_type to "fast" in your Search API request body and you are using the new perplexity fast search api tier — announced by Perplexity on 24 September 2026, it returns 95 per cent of search results in 230 milliseconds or less and is priced in the official API docs at 1 dollar per 1,000 requests. It runs on Photon, Perplexity's new Rust-based retrieval and ranking engine, which the company says was built by a small team of engineers working with hundreds of AI agents. If you build research agents, SEO tooling or anything that hammers a web-search endpoint in a loop, this is the change that makes those loops feel instant instead of sluggish.

📺 Watch: Perplexity Just Made AI Search 68% Cheaper

🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside · Want AI SEO help 1-on-1? Book a free SEO strategy session →

What The Perplexity Fast Search API Actually Is

Perplexity's Search API returns real-time, ranked web results as structured data — titles, URLs, snippets, dates and last-updated stamps — rather than a written answer. Fast Search is a new speed tier inside that same API: same endpoint, same request shape, one extra parameter. According to the official announcement on the Perplexity API platform forum (24 September 2026), the headline numbers come from Photon, the retrieval and ranking service Perplexity built from scratch in Rust to replace its previous system.

The pitch is aimed squarely at agent builders. A human waiting on a chat answer barely notices 800 milliseconds; an agent that fires dozens of searches per task absolutely does. Cutting per-search latency compounds across every step of an agentic workflow, which is why fast, cheap search endpoints are becoming the quiet battleground of the agent stack — the same logic that makes routing layers like the OpenRouter Fusion API matter.

Perplexity Fast Search API Pricing And Speed

The two numbers that matter, both from Perplexity's own materials fetched at the time of writing:

• Speed: 95 per cent of Fast Search results return in 230 milliseconds or less, per the official announcement.
• Price: the official API documentation lists Fast Search at 1 dollar per 1,000 requests.

One billing detail worth knowing: the Search API accepts up to 5 queries in a single request, and a multi-query request counts as one billing unit while consuming one rate-limit unit per query. Batch your related lookups and the effective cost per query drops further — five queries for the price of one request is the cheapest way to run broad research sweeps.

If you want ready-made agent workflows that turn cheap search calls into content, rankings and revenue, check out the AI Profit Boardroom — members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls: join the AI Profit Boardroom. Want a 1-on-1 plan for your AI SEO stack first? Book a free SEO strategy session.

How To Use The Perplexity Fast Search API

The mechanics, straight from the official quickstart documentation:

1. Endpoint: send a POST request to api.perplexity.ai/search with your API key as a bearer token and a JSON body.
2. Switch on the fast tier: add the search_type parameter with the value "fast" to the request body. That single parameter is the whole migration.
3. Shape the results: query accepts a single string or an array of up to 5 queries; max_results takes 1 to 20 (default 10); search_context_size accepts low, medium or high to control how much page content is extracted.
4. Filter the web: search_domain_filter allowlists or denylists up to 20 domains, search_language_filter takes up to 10 language codes, and a country parameter localises results by ISO country code.
5. Cap the spend: max_tokens and max_tokens_per_page set manual extraction budgets, up to 1,000,000 tokens.

The response is a JSON object with a results array — each entry carrying title, URL, snippet, date and last_updated — plus a request id. Because it is structured data rather than prose, it drops cleanly into whatever your agent does next, whether that is a summarisation step in Claude Code or a research chain like the setups in our best AI agent for research round-up.

What 1 Dollar Per 1,000 Requests Buys In Practice

Flat, predictable pricing is easy to plan around, so it is worth doing the arithmetic before you design the workflow. At the documented rate of 1 dollar per 1,000 Fast Search requests, a monitoring agent that runs 1,000 searches a day costs about 1 dollar a day — roughly 30 dollars a month for a volume of querying that would have been a genuine line item on older answer-style endpoints. A weekly content-gap sweep of 200 queries costs about 20 cents. A one-off research project that fires 5,000 searches costs 5 dollars.

Batching improves that again. Because a single request carries up to 5 queries and bills as one unit, a disciplined workflow that groups related lookups gets up to 5,000 queries per dollar. The trade-off sits in rate limits rather than billing — each query inside a batched request still consumes its own rate-limit unit — so batching saves money without magically raising your throughput ceiling. For most automation builders, the practical conclusion of the perplexity fast search api pricing is simple: search volume stops being the thing you optimise first, and result quality per query takes over as the real constraint.

Setting Fast Search Up Inside An Agent Workflow

A sensible integration pattern looks like this. Wrap the search endpoint as a single reusable tool in whatever framework you run — a custom tool in your agent harness, an HTTP node in an n8n workflow, or a plain function in a script — and make search_type "fast" the default inside the wrapper so every caller gets the fast tier without thinking about it. Keep the API key in an environment variable, not in the workflow definition.

Then add two cheap disciplines. Cache results keyed on the query string with a sensible expiry, so repeated runs of the same job do not re-buy identical results. And log the request id that comes back with every response next to the decision your agent made with it — when an automated pipeline produces a strange output three days later, that id is the difference between a five-minute diagnosis and an afternoon of guessing.

📺 Watch: NEW Perplexity Computer Updates are WILD! 🤯

Where Fast Search Fits In An Automation Stack

Three placements make immediate sense. First, research agents: anything that loops search-read-decide gets a direct speed-up, including Hermes setups that lean on Hermes agent web search patterns. Second, SEO tooling: rank tracking, SERP sampling and content-gap sweeps are exactly the high-volume, low-stakes queries a 1-dollar-per-1,000 tier is made for — the strategies in the Perplexity SEO guide get materially cheaper to run at scale. Third, monitoring: freshness checks and brand mentions that were too expensive to poll hourly stop being a cost conversation at this price point.

If you are systematising this properly, wire Fast Search in as a standard tool inside your Agent OS playbooks rather than hard-coding it per project — and when you are choosing which model consumes those search results, the Goldie Bench write-up covers how the current agent brains compare in hands-on tests, so you can pair a cheap search layer with a model that actually reasons over it well.

The Caveats Before You Migrate Everything

• Quality trade-offs are workload-specific. The announcement leads with latency, not ranking depth. Perplexity keeps the standard tier alongside Fast Search, which tells you both have a job — test your own queries on both before switching production traffic.
• It returns results, not answers. If you currently use Perplexity's answer-style endpoints, Fast Search is a different shape of product: your agent has to read and synthesise the results itself.
• Token extraction is a separate lever. A high search_context_size pulls more page content into each result; if you are optimising for speed and cost, keep it low and fetch depth only when a result earns it.
• SDK support may lag. The docs note workarounds for setting search_type through some SDKs — the raw HTTP call is the reliable path on day one.

Verdict: Who Should Switch To Fast Search?

Switch anything high-volume and latency-sensitive now: agent research loops, SEO sweeps, monitoring jobs. The perplexity fast search api gives you a documented 230-millisecond 95th-percentile response and a flat 1-dollar-per-1,000-requests price, and turning it on is a one-parameter change you can revert just as fast. Keep judgement-heavy, one-off research on the standard tier until your own side-by-side tests say otherwise. If your searches feed money pages, the workflows in how to make money with Perplexity pair naturally with this cheaper, faster backend — the economics of running those systems at scale just improved.

If you want the full system for turning AI search into traffic and income — the Agent OS bonus, 1,000+ done-for-you workflows, daily tutorials and weekly coaching calls — check out the AI Profit Boardroom. Prefer to map it 1-on-1 for your business first? Book a free SEO strategy session.

Real wins from inside the AI Profit Boardroom

See all 3,000+ members →

← Back to all posts