Perplexity’s public docs describe a developer API platform, not a consumer Pro membership page. The overview positions Router, Agent, Search, and Embeddings APIs for real-time, web-wide research and Q&A inside your products. Quickstart copy states Pay-as-you-go pricing for all APIs with no subscription required, and Agent docs stress web-grounded answers with built-in citations plus multi-provider model access through one interface.
Choose Perplexity when you need programmable web research: Agent calls that return cited answers, Search API raw ranked web results for your own pipeline, Router access to open-weight models on an OpenAI-compatible path, or Embeddings for semantic search and RAG. Expect separate billable layers—model tokens, Agent API tool invocation fees such as web_search, Search API per-1K request fees, and Decisions API input-token billing—rather than one flat consumer plan in these pages. Skip these docs if you only want a browser chat membership; the consumer homepage was not available here, and these sources do not publish a public web subscription price. Prefer Consensus or Elicit when the job is paper-centric literature review instead of general web grounding.
A good fit for
Product teams that need web-grounded answers with built-in citations inside an app or agent workflow.
Builders who want ranked web search results or embeddings without wrapping an LLM in the same call.
Teams consolidating multi-provider models behind one Perplexity API key with pay-as-you-go metering.
Know the trade-offs
Agent tool fees, Search API per-1K prices, and model token rates are separate meters—budget them independently.
These docs do not publish a consumer web subscription tier; do not treat API line items as Pro monthly dues.
Free output tokens on Decisions models do not prove every other model or tool is included at no charge.
What the Perplexity APIs do.
Agent API for cited, web-grounded answers
The Agent API is documented as a multi-provider interface for LLM apps: send a message and receive web-grounded answers with built-in citations in one call, keep conversation context across turns, and optionally reach OpenAI, Anthropic, Google, xAI, and other providers through a single key. Docs highlight presets for a researched answer in one shot, or granular control over model, reasoning, token budgets, and tools. The endpoint is POST https://api.perplexity.ai/v1/agent, with /v1/responses as an OpenAI-compatible alias. Tools are billed apart from model tokens. Pricing lists web_search (standard) and Fast Search, image_search, fetch_url, people_search, finance_search, and a sandbox session for code. A chatty agent that searches every turn accrues tool charges even when token rates look modest. Official Python and TypeScript SDKs are recommended; authentication uses PERPLEXITY_API_KEY or an explicit key.
Quickstart guidance separates jobs clearly. Use Search when you need Search API raw ranked web results without LLM processing—for custom pipelines, indexing, analysis, or training—with filtering, multi-query support, and domain controls. Use Router for open-weight models hosted by Perplexity through one OpenAI- or Anthropic-compatible endpoint with routing and failover. Use Embeddings for semantic search and RAG vectors. Decisions answers yes/no, multiple-choice, and scored questions about text and images with calibrated probabilities. Router usage is billed per token at each model’s published rates with no per-request fees on that overview. Search API Pricing lists a per-1K schedule distinct from Agent tools. Decisions billing is input-token only for listed decider models, with output tokens marked free and no per-request fee on those rows. Keep those meters separate; mixing Search batches with Agent tool loops under one “search fee” misstates the bill.
Across quickstart and Agent pages, Perplexity repeats Pay-as-you-go pricing for all APIs and states no subscription required for that API access path. Pricing docs point admins to billing setup, payment methods, and usage monitoring in Admin, and to Rate Limits & Usage Tiers for throughput. Purchase options include API credits through AWS Marketplace and contacting sales for enterprise talks. Getting started means generating an API key on the API Keys page in the API Console and making a first call in under three minutes, with REST and SDK paths plus streaming. Transparent Pricing on the Agent page means seeing exact token counts and published costs per request—not a bundled consumer membership in these developer docs. If you need a named monthly Pro card on a marketing site, these pages will not supply that SKU.
Product teams can wire the Agent API so users ask questions and receive cited, web-grounded answers without building a separate crawler. Presets reduce setup for a researched reply in one call; tools add web or finance search when the prompt needs live sources. This path fits AI assistants, research Q&A widgets, and agentic workflows that must show where claims came from. Compare latency and tool spend against Search plus your own model if you need tighter generation control.
Feed a custom RAG or ranking pipeline
When you already own the LLM layer, Search API raw ranked web results and the Embeddings API give retrieval ingredients without forcing Perplexity’s answer prose. Filter domains, run multi-query retrieval, then embed passages for your vector index. Data teams use this for indexing, analysis, or training sets where the deliverable is structured hits or vectors. Keep the Search per-1K schedule in the forecast so batch crawls do not surprise finance.
Route open-weight and third-party models from one key
Router consolidates open-weight models hosted by Perplexity behind one endpoint and key, which helps when you already have OpenAI or Anthropic SDK code and only need a base-URL change. Agent adds multi-provider access with web tools when the job needs grounding plus a specific vendor model. Platform groups adopt this to reduce per-provider accounts while metering tokens at each model’s published rates. Suggest Perplexity here when operational simplicity matters more than a single upstream vendor console.
How to get started with Perplexity
1
Create a project API key
Open the API Console, generate a key on the API Keys page, and store it as PERPLEXITY_API_KEY or pass it explicitly in your client.
2
Pick Agent, Search, Router, or Embeddings
Match the job: cited web answers and tools on Agent, raw ranked hits on Search, open-weight OpenAI-compatible calls on Router, vectors on Embeddings.
3
Install an SDK or call REST
Install perplexityai for Python or @perplexity-ai/perplexity_ai for TypeScript, or post to the documented REST endpoints with streaming if you need incremental output.
4
Read pricing meters before scale
Check model token rates, Agent tool fees, and Search API per-1K prices as separate lines, then monitor usage from Admin analytics.
Docs describe Pay-as-you-go pricing for all APIs with no subscription required. Agent tools include web_search (standard) at $0.0025 per invocation ($2.50 per 1,000 invocations) and web_search (Fast Search) at $0.001 per invocation ($1.00 per 1,000 invocations). Search API Pricing lists Search API at $5.00 per 1K requests and Search API with Fast Search at $1.00 per 1K. Decisions rows show pplx-decider models at $0.02 per 1M input tokens with free output tokens; a sample notes a request with 600 input tokens costs $0.000012. Router is per-token at each model’s published rates with no per-request fees on that overview.
A consumer web subscription price was not verified on these developer pages. A named Pro monthly membership was not verified here. Do not map Agent tool fees, Search API per-1K charges, or model token rates into one invented plan. Third-party Agent model rate cards sit on separate model pages not fully reproduced here.
Your workflow is systematic review and paper screening.
Elicit is built around research workflows over papers and evidence tables. Perplexity’s docs instead center API access for live web answers, ranked search results, and embeddings you embed into your own stack.
You want a conversational search product rather than API meters.
Andi presents an end-user conversational search experience. Perplexity here is evaluated as a pay-as-you-go API platform with Agent tools, Search per-1K pricing, and Router models for developers integrating research into software.
Perplexity: common questions
Is Perplexity in these docs a consumer chat subscription?
No. The listed sources are developer docs for Router, Agent, Search, Embeddings, and related APIs. They repeat pay-as-you-go API pricing and do not publish a public consumer web subscription price.
Docs describe web-grounded answers with built-in citations in one call, optional conversation context, multi-provider models, presets, and tools such as web and finance search through one interface.
When should I use the Search API instead of Agent?
Use Search when you need raw ranked web results without LLM processing—for custom AI workflows, indexing, analysis, or training—rather than a cited natural-language answer from Agent.
Tool costs are separate from model token costs. Pricing lists per-invocation fees for tools such as web_search, image_search, and others, plus sandbox session charges, on top of whatever the chosen model bills per token.
The pricing overview states Router API usage is billed per token at each model’s published rates and that there are no per-request fees for that API on that page.
NavHunt did not run Perplexity. Metered API prices come from the public developer documentation for the overview, pricing, quickstart, and Agent API, reviewed on 6 October 2026. This description is based on Perplexity's official materials reviewed on 2026-10-06, including Overview - Perplexity, Pricing - Perplexity, Quickstart - Perplexity, Agent API - Perplexity.