LIVE TICKER
[Live wire]Connecting to the newsroom…
Verified pricing · checked 29 October 2026

Perplexity API Alternatives: Retrieval and Answer Costs Compared

Perplexity splits search from generation. This comparison keeps that split visible, using vendor prices checked on 29 October 2026.

perplexity.ai•5 alternatives•6 vendor sources linked

Perplexity’s developer platform now sells several products under one account. The Search API returns ranked web results as JSON. The Agent API runs hosted models with tools. An Embeddings API covers vectors, and a Decisions API answers scored questions.

That structure matters when you shop for alternatives. A cheap per-request search price does not include a written answer, and an answer API does not always expose raw results. Comparing the two as if they were the same product produces bad decisions.

The prices below separate retrieval from generation. Every number is copied from the vendor’s own documentation, and each row links to the page it came from.

What the Perplexity API charges today

VendorPlanPriceWhat it includes
PerplexitySearch API, standard$5 per 1,000 requestsRanked results with extracted text. Up to 20 results per query, five queries per request.
PerplexitySearch API, fast$1 per 1,000 requestsLower-latency results in the same response format.
PerplexityAgent API web search tool$2.50 per 1,000 invocationsStandard web search inside an agent run. Fast search costs $1 per 1,000.
PerplexityAgent API modelsProvider ratesThird-party models are billed at each provider’s published token rates, plus tool costs.
PerplexityEmbeddings, standard$0.004–$0.03 per 1M tokens0.6B model at $0.004, 4B model at $0.03. Contextualised versions cost $0.008 and $0.05.
PerplexityCode sandbox$0.03 per sessionEach session is billed on a 20-minute window, not a runtime cap.

Taken from the Perplexity API pricing page. Search API billing counts successful requests, not queries inside a request. Source

Five alternatives, split by what they deliver

Three of these can return a finished answer, and two return retrieval only. The price column shows which is which.

Tavily

tavily.com

Pricing page →

Best for: Agent loops that want Markdown context and an optional short answer.

Free 1,000 credits per month. Pay as you go $0.008 per credit, or $30 per month for 4,000 credits.

  • +An include_answer option returns a short synthesised answer with the results.
  • +Domain allow and deny lists are supported per request.
  • +Framework integrations exist for common agent libraries.
  • −Credit cost varies by request type, so the headline rate is not per search.
  • −Requests stop when the monthly credit balance is exhausted.

Best for: Semantic retrieval and full page contents for RAG.

Lite $4, Auto $7, Deep $12 per 1,000 requests. Contents $1 per 1,000 pages. Answer $5 per 1,000.

  • +Answer endpoint is priced separately, so you can switch generation off.
  • +Contents endpoint returns full page text per 1,000 pages.
  • +Free tier includes $10 of credits every month.
  • −Every extra result above ten adds $1 per 1,000 requests.
  • −Answer quality depends on the retrieved page set, which you must tune.

Brave Search API

brave.com

Pricing page →

Best for: Grounding on an independent index with a citation-friendly Answers plan.

Search $5 per 1,000 requests. Answers $4 per 1,000 requests plus $5 per million input and output tokens.

  • +No fallback to Google or Bing, which Brave states explicitly.
  • +Answers plan supports streaming and is OpenAI SDK compatible.
  • +Includes $5 of free credits every month.
  • −Answers is capped at 2 queries per second.
  • −Token billing makes the final cost depend on answer length.

OpenAI web search

openai.com

Pricing page →

Best for: Teams already running OpenAI models that want search inside the same call.

Web search $10 per 1,000 calls, plus search content tokens at model rates. Non-reasoning preview tool costs $25 per 1,000.

  • +Search results flow straight into the model response, with no second bill.
  • +Cheapest current model, gpt-6-luna, is listed at $0.10 input and $0.50 output per 1M tokens.
  • +The same key covers chat, search, file search and containers.
  • −You cannot buy retrieval without also paying for tokens.
  • −Search content tokens are billed in addition to the call fee.

Serper

serper.dev

Pricing page →

Best for: Pure Google SERP retrieval at the lowest published unit price.

$50 for 50,000 credits ($1 per 1,000), down to $0.30 per 1,000 at 12.5 million credits.

  • +Top-up model with no monthly subscription.
  • +2,500 free queries on sign-up, no credit card required.
  • +High default concurrency on Ultimate credits.
  • −No generated answer: you supply the model and the prompt.
  • −Results mirror Google, including its ranking behaviour.

What does a Perplexity API request actually cost?

A standard Search API request costs $0.005, which is $5 per 1,000. That request can carry up to five queries, and it is still billed as one unit. Invalid, rate-limited and failed requests are not billed at all.

The fast mode costs $0.001 per request. It returns the same response format with lower latency, so it suits autocomplete and voice agents where speed matters more than depth.

Neither price includes a written answer. If you want prose with citations, you either call the Agent API or run your own model over the returned snippets. The Agent API adds tool costs such as $2.50 per 1,000 web searches, plus the model’s own token bill.

That split is why some teams move to Tavily. A single call can return results and a short answer, which removes one integration step and one bill. The trade-off is less control over which model writes the answer.

When is Perplexity the right choice?

Perplexity is strongest when you want ranked SERP-style results with page text already extracted. The pricing is transparent, five queries share a billing unit, and the extracted text length does not change the fee.

It is weaker when you need full-site crawling. The Search API returns results, not an entire domain. Firecrawl and similar crawlers are the right tool for that job, and they are priced per page or per credit.

It is also weaker when your finance team wants one predictable invoice. Search fees, tool fees and model tokens arrive as separate line items, and the total depends on how your agent behaves at runtime.

How to compare answers fairly

Answer quality is not priced. Two providers can charge the same $5 per 1,000 requests and return very different snippets. The only reliable test is your own query set with your own evaluation criteria.

Build a set of fifty real questions, then score each provider on whether the cited page contains the answer. Track cost per correct answer rather than cost per request. That metric exposes providers that look cheap but require two calls to be useful.

Check the billing rules too. Perplexity bills a successful response even when it returns no results, while Serper only deducts credits for successful responses. Firecrawl does not charge for a scrape that returns nothing, but it does charge one credit for a 403 or 404 page.

Frequently asked questions

Is the Perplexity API the same as Perplexity Pro?+

No. Perplexity Pro is the consumer subscription. The API is billed separately through the developer platform, with its own keys, rate limits and usage tiers.

Which alternative is cheapest for retrieval only?+

Serper lists $1 per 1,000 queries on a $50 top-up, falling to $0.30 at the largest pack. Perplexity’s own fast search is $1 per 1,000 requests, and Exa Lite is $4 per 1,000.

Can I get a cited answer without running a model?+

Yes. Tavily offers an answer option alongside results, Exa sells an Answer endpoint at $5 per 1,000, and Brave Answers charges $4 per 1,000 requests plus $5 per million tokens.

Does Perplexity still sell Sonar?+

The documentation now lists Search, Agent, Embeddings and Decisions APIs, and includes a migration guide for Sonar. Check the current docs before you write new code against the old model names.

Sources

Every price on this page was read from these pages on 29 October 2026. Vendors change prices without notice.

Related reading