Perplexity API 定价详解：Search、Sonar、密钥与速率限制

Perplexity API pricing starts at $5 per 1,000 successful standard Search requests or $1 per 1,000 Fast Search requests; Agent adds model-token charges and tool charges.

With a funded key in PERPLEXITY_API_KEY, call POST https://api.perplexity.ai/search to retrieve results rather than generate an answer.

Output:

Install the HTTP dependency with python -m pip install requests.

Perplexity API pricing at a glance

Search returns ranked web results as structured data, while Agent is the current interface for generated answers and tool use.

Search, Agent and Sonar pricing

Search API request prices

A request can contain up to five queries for a single billable request. However, a five-query request consumes five rate-limit units. Track billable requests and query-unit capacity separately when estimating a batched workload.

The billing rule is explicit: "A successful response is billed even when it returns no results." Invalid, rate-limited and upstream-failed requests are not billed.

Search's max_tokens controls returned webpage-content tokens, not answer generation. If your application needs synthesis, budget for Agent or a separate model step: Search does not generate an answer.

Current Agent Sonar model and tool prices

These token rates apply to perplexity/sonar in Agent API, not to a historical Sonar Chat Completions price table.

The listed cache-read rate does not mean every request receives cached-input pricing. A standalone Search request and an Agent search-tool invocation are different billing units.

Estimate your workload cost

This calculator separates Search billing from Agent billing using the Search and tool rate card and Agent Sonar model rates. Edit the usage dictionary with successful Search request counts, token counts and tool invocation counts. Keep ordinary input tokens and cache-read tokens separate to avoid counting the same tokens twice.

Output:

Tested Oct 22, 2026.

Was Sonar discontinued?

Sonar Chat Completions support ended September 27, 2026. Synchronous and streaming requests continue through gradual Agent API reformulation; asynchronous Sonar requests are no longer supported.

For new integrations, use POST https://api.perplexity.ai/v1/agent. Migrate asynchronous workloads to Agent background mode and retrieve results with GET /v1/agent/{id}. Continuing synchronous compatibility is not the same as continued support for the legacy interface.

API key and free credits

Create a project in the Perplexity API Console, open API keys and choose + Generate API Key. Save the key securely: its full value is shown only once.

Add prepaid credits through Billing, then set PERPLEXITY_API_KEY in the environment that launches Python. Keep the key out of browser code and public repositories; exhausting prepaid credits blocks API keys.

Does Perplexity Pro include free API usage?

The official help center states "No complimentary API credits". Treat consumer subscriptions and API usage as separate budget items.

Rate limits

Search allows 50 query units per second and a 50-unit burst. Agent Tier 0 allows 1 QPS shared across models and organization API keys; status polling consumes that same allowance. Leave capacity for new requests rather than spending the entire allowance on polling.

Python libraries

The HTTP examples here use requests so you can see the request body, authentication and throttling response directly.

Batch scholarly queries and build a paper list

Search supports domain filtering with up to 20 domains per request. This example batches queries, restricts retrieval to the chosen domains, removes duplicate URLs and handles missing publication dates.

Output:

Use an allowlist or a denylist, not both in the same request.

These examples stop on HTTP 429 throttling and print Retry-After when supplied. Wait for that interval before resubmitting; if it is absent, back off rather than retrying immediately.

Common errors

Agent-only fields such as model, preset, instructions, tools and max_output_tokens are not part of the Search request schema. Keep retrieval and generation request bodies separate.

Terms and compliance

Perplexity documents that it "does not use Customer Content to train generative AI models," while noting that product-specific terms can apply.

Do not turn that into a zero-retention promise. "Zero Data Retention is not default": eligibility, account configuration, a written agreement and model support matter. Third-party model terms also apply.

Covered Search services have a product-specific addendum that overrides the underlying agreement for those services. Read it alongside the API terms.

The Search addendum permits retention and reuse of Search inputs and outputs for lawful business purposes, including product development. Do not submit personal data without Perplexity's explicit written authorization. Receiving Search output does not give you ownership of that content.

Review service terms and source licenses before storing or redistributing content. OpenAlex's public-domain license applies to its data; do not automatically extend that permission to publisher PDFs linked from records.

Perplexity alternatives for research

Perplexity Search and Agent are built for the open web. If your product needs scholarly papers, the right alternative depends on what you need back.

Where Perplexity is the better choice:

for broad web context, current events or general questions, Perplexity covers the open web and SciSpace does not. Perplexity's Agent API can also return files such as CSV, PDF, Word, PowerPoint and Excel through its code sandbox.

Lost Sonar's academic mode? The SciSpace API

Sonar's search_mode=academic is gone, and the Agent API has no direct replacement. The migration guide suggests domain filters and prompting instead, as the paper-list example above does. That gets you web pages from scholarly domains, not paper records: no consistent DOIs, journals or study types, and no screening, extraction or reference export.

The SciSpace API puts the full SciSpace Agent behind one call. It searches, reads, analyses and writes, and delivers the result as ready-to-use files.

It comes as two APIs:

What a research product gets that a web search API doesn't

• Beyond papers. One task can also search 11 national and regional clinical-trial registries plus the WHO registry, Google Patents, grant databases and 77 public science databases such as PubChem and UniProt.
• Finished files. One run can return a Word report, a LaTeX PDF with its BibTeX, a PowerPoint deck or plots. You get links to every file it created.
• Follow-ups in the same thread to refine, extend or reformat the work.
  

Frequently asked questions

How is Perplexity API pricing calculated?

Search is billed per successful request, with separate standard and Fast Search rates and no additional token charge. Agent API bills model tokens plus applicable tool charges.

Does Perplexity Pro include free API credits?

The official help center says there are no complimentary API credits. Do not treat a consumer subscription as an API credit allowance.

Was the Sonar API discontinued?

Sonar Chat Completions support has ended. Synchronous and streaming requests continue through gradual Agent API reformulation, but asynchronous Sonar requests are no longer supported. Migrate background workloads to Agent API.

Does batching Search queries reduce rate-limit usage?

No. Each query in the array still consumes query-unit capacity. Track billing requests and rate-limit units separately.

Is there an official Perplexity Python SDK?

Yes. Install it with `pip install perplexityai` and import `Perplexity` from `perplexity`. The official SDK reads the `PERPLEXITY_API_KEY` environment variable.

Does the Perplexity Search API generate cited answers?

No. Search returns ranked retrieved results rather than a generated answer. Use Agent API with web search for synthesis, or pass Search results to your own model.

Can I keep academic search mode after migrating from Sonar?

No. There's no direct Agent API equivalent to the legacy search_mode=academic; the migration guide recommends web_search with domain filters and prompting instead [link]. If you need paper records rather than web pages, use a scholarly API: OpenAlex for open metadata, or the SciSpace Search API for papers with full metadata from SciSpace's 280M+ index, Google Scholar, PubMed and arXiv (early access, $5 free credits at launch).

 What's the best Perplexity API alternative for academic research? It depends on what you need back. For open scholarly metadata, OpenAlex is free with a daily allowance. For papers with full metadata in one plain-language query, plus screening, extraction tables and reports with BibTeX, the SciSpace API is in early access with $5 in free credits at launch. For general web context, Perplexity remains the better fit.

Last checked Facts, prices and code examples in this guide were checked on October 22, 2026.

Sources

1. Perplexity API pricing
2. Agent API model rates
3. Perplexity Search overview
4. API key management
5. Rate limits and usage tiers
6. Search quickstart
7. Search API reference
8. Sonar migration overview
9. Agent API quickstart
10. Agent API background mode
11. Perplexity projects and prepaid credits
12. Which Perplexity subscription plan is right for you?
13. Official Perplexity SDK overview
14. Search domain filters
15. Perplexity privacy and security
16. Perplexity API Terms of Service
17. Perplexity Search API terms addendum
18. OpenAlex access and pricing
19. OpenAlex snapshot download tutorial
20. SerpApi Google Scholar connector
21. SerpApi pricing
22. Migrating Sonar features to Agent API
23. Installation of Requests

注意转载授权 · 阅读原文