TL;DR
- Search API: $5 per 1,000 requests standard, $1 with Fast Search. No token charges. Up to 20 results per request.
- Agent API: model token rates plus tool fees, e.g.
web_searchat $2.50 per 1,000 calls. - A query array of five queries counts as one request. That's the biggest money-saver here.
- No free API tier is listed. Public SimpleQA: 85.92% in Tavily's open repo.
“How much does the Perplexity API cost?” is one of those questions with a frustrating answer: which one?
Perplexity now sells several APIs, and they’re billed in completely different ways. The Search API is billed per request with no tokens at all. The Agent API is billed per token and per tool call. Mix them up and you’ll either overpay or wildly misread your forecast.
Here’s the whole price list, how each part is billed, and the one rule that can cut your search bill by 5×. All numbers are from Perplexity’s pricing docs, checked November 2, 2026.
The Perplexity APIs and what they cost
| API | What you get | Price |
|---|---|---|
| Search API (standard) | Ranked web results, page text in snippets, no model | $5 per 1,000 requests |
| Search API (Fast Search) | Same format, lower latency (search_type: "fast") | $1 per 1,000 requests |
| Agent API | Third-party models with Perplexity’s tools | Model token rates + tool fees |
↳ web_search tool | Standard web search inside an agent run | $2.50 per 1,000 calls |
↳ web_search (fast) | Fast Search inside an agent run | $1.00 per 1,000 calls |
↳ fetch_url tool | Fetch and extract a given URL | $0.50 per 1,000 calls |
↳ people_search, finance_search | Specialised lookups | $5 per 1,000 calls |
↳ sandbox | Code execution session (up to 20 min billing window) | $0.03 per session |
| Orchestrator API | Perplexity’s routing across models | Token rates per model, no request fee |
| Embeddings | pplx-embed-v1 models | $0.004–0.05 per 1M tokens |
Perplexity’s docs include a migration guide away from the older Sonar models. If your code or your budget spreadsheet still says Sonar, read that guide before you forecast anything.
How the Search API is billed
Three rules, and they’re all good news:
- Per successful request, not per result and not per token. Up to 20 results per request at the same price.
- One request can hold several queries. A query array of five queries in one
POST /searchcounts as one request: $0.005 standard, or $0.001 fast. - Text length doesn’t change the price.
max_tokensandmax_tokens_per_pagecontrol how much extracted text you get, at no extra cost.
Perplexity’s own example: 10,000 successful standard requests in a month cost 10 × $5 = $50.
Say your agent fans every question out into five searches, and you handle 20,000 questions a month. Sent one at a time, that’s 100,000 standard requests: $500. Sent as one five-query array per question, it’s 20,000 requests: $100. In Fast mode, $20. Same searches, same results.
The cheapest Perplexity search is the one you batched.
How much does the Perplexity API cost at volume?
| Searches per month | Standard ($5/1k) | Fast ($1/1k) |
|---|---|---|
| 10,000 | $50 | $10 |
| 100,000 | $500 | $100 |
| 1,000,000 | $5,000 | $1,000 |
To compare with every other search API at your volume, use the price calculator.
Search API vs Agent API: which do you need?
You have your own model
Search API
Results and page text as JSON, no tokens billed. Feed them to whatever model you like.
You want Perplexity to run the agent
Agent API
Pick a third-party model, let it call web_search, fetch_url and friends. You pay tokens plus tool fees.
If you already have a model and a prompt, the Search API is almost always the cheaper path, because you’re not paying a second margin on tokens. The Agent API earns its keep when you’d otherwise build the tool-calling loop yourself.
Can I use the Perplexity API for free?
Perplexity’s pricing page doesn’t list a free API tier. To prototype for free, Parallel (5,000 requests a month), Tavily (1,000 credits a month) and Brave ($5 of credits a month) all have one.
Is the Perplexity Search API good value?
At $1 per 1,000 in Fast mode, it’s one of the cheaper ways to get results with page text and no token bill: the same list price as Parallel’s fast mode. At $5 standard, it’s priced like Brave, You.com and Linkup.
On accuracy, the public number is from a competitor: Tavily’s open SimpleQA repo scored Perplexity Search at 85.92%, below Tavily’s own 93.3% in the same run. Perplexity’s own figure, 85.8 F-score for Sonar Pro, has no public results. See the SimpleQA table.
Perplexity API alternatives
| Need | Alternative | Price per 1,000 |
|---|---|---|
| Lowest price | Keirolabs lite | $0.10 |
| Same price, own index | Parallel fast | $1 |
| Higher verified SimpleQA | You.com | $5 |
| Neural search with page text | Exa | $4–7 |
More in best web search APIs and Tavily alternatives.
FAQ
Does Perplexity have API access?
Yes: the Search API, Agent API, Orchestrator API and Embeddings, all billed through the Perplexity API platform.
Is the Perplexity API the same as Perplexity Pro?
No. A Pro subscription covers the consumer app. API usage is billed separately, per request or per token.
Why is my Perplexity API bill higher than expected?
Usually because the Agent API’s model tokens were counted on top of tool calls. If you only need results, the Search API has no token charges at all. And check whether you’re batching queries.