The Perplexity API Platform is the developer product of Perplexity, separate from its consumer and Enterprise subscriptions. A Perplexity subscription is not required to buy API credits or use the API.[9] The platform is metered rather than seat-based: a customer buys credits in advance and API calls draw them down at published per-token, per-invocation or per-request rates. This article sets out the rate card as documented on 2026-10-13, the tier system that governs rate limits, the project and billing model, and the AWS Marketplace route.
Editions
The pricing page groups the platform into several APIs.[1]
- Orchestrator API. Billed per token at each model’s published rates, with no per-request fees. Every model has its own input and output rate, cache reads are billed at the model’s discounted cache-read rate, and reasoning tokens are billed at the output rate. The customer is billed at the requested model’s rates regardless of how the request is served.[1] The model catalog page states that the Orchestrator API is in private preview and access is requested from Perplexity.[8]
- Agent API. A multi-provider interface to third-party models, with token-based pricing at each model’s published rates and integrated tools. The pricing page names OpenAI, Anthropic, Google, xAI, Z.AI, Moonshot AI and NVIDIA as providers.[1]
- Search API. Raw web search results with filtering.[1]
- Embeddings API. Standard and contextualized embedding models.[1]
- Decisions API. Calibrated yes, no, multiple-choice and scored answers about text and images.[1]
- Sonar Chat Completions. The documentation states that Sonar Chat Completions support ended on September 27, 2026. Synchronous and streaming requests keep working while being reformulated as Agent API requests, rolling out gradually by model; asynchronous Sonar requests are not reformulated and are no longer supported, and users are told to move them to background mode.[6] Perplexity recommends migrating existing Sonar usage to the Agent API and using the Agent API for all new projects.[6]
Metrics
Agent API tokens and tools
Model usage and tool usage are charged separately. Tool costs are separate from model token costs.[1]
| Tool | Price | Basis |
|---|---|---|
| web_search (standard) | $0.0025 per invocation | $2.50 per 1,000 invocations[1] |
| web_search (Fast Search) | $0.001 per invocation | $1.00 per 1,000 invocations[1] |
| image_search | $0.0025 per successful invocation | Filters for domain, format and safe search[1] |
| fetch_url | $0.0005 per invocation | Fetches and extracts content from specific URLs[1] |
| people_search | $0.005 per invocation | $5 per 1,000[1] |
| finance_search | $0.005 per invocation | $5 per 1,000[1] |
| sandbox | $0.03 per session | A session covers up to 20 minutes of active use for billing purposes; this is a billing window, not a runtime cap[1] |
SDK search queries made from inside the sandbox are billed at $0.0025 per request, the same as web_search.[1] The pricing page’s worked example prices a 500 input and 200 output token call with one web search at $0.0059 in total, and notes that when present on a completed response, usage.cost.total_cost reports the calculated request cost.[1] The Agent API also offers presets named fast, low, medium, high and xhigh; a preset called by name resolves to the latest Perplexity-recommended configuration, and presets are not pinned to a version, so cost can drift unless a frozen configuration is copied into the request.[7]
Search API requests
The Search API costs $5.00 per 1,000 requests, or $1.00 per 1,000 with Fast Search. The billing unit is each successful POST /search request, not each query in the request, so a successful request containing up to five queries is one billing unit. Invalid requests, rate-limited requests and upstream failures are not billed, a successful response is billed even if it returns no results, and there are no token-based charges.[1]
Embeddings and Decisions
Standard embeddings: pplx-embed-v1-0.6b (1024 dimensions) at $0.004 per million tokens and pplx-embed-v1-4b (2560 dimensions) at $0.03. Contextualized embeddings: pplx-embed-context-v1-0.6b at $0.008 and pplx-embed-context-v1-4b at $0.05.[1] The Decisions API charges $0.02 per million input tokens for both listed models, output tokens are free and there is no per-request fee; the state, every question and any images count as input.[1]
Counting / floors
Usage tiers
Usage tiers determine rate limits and access to beta features based on cumulative API spending. They advance automatically as credits are purchased, and a reached tier is kept permanently with no downgrade.[2]
| Tier | Total credits purchased | Agent API limit |
|---|---|---|
| Tier 0 | $0 | 1 QPS |
| Tier 1 | $50+ | 3 QPS |
| Tier 2 | $250+ | 8 QPS |
| Tier 3 | $500+ | 17 QPS |
| Tier 4 | $1,000+ | 33 QPS |
| Tier 5 | $5,000+ | 33 QPS |
Tiers are based on cumulative purchases across the account lifetime, not on the current balance.[2] The Agent API limit applies per API organization, is shared across models and API keys, and uses a rolling one-second window.[2] Orchestrator API requests that are rate limited or hit an overloaded upstream model return a 429 with a Retry-After header, and requests rejected with a 429 are not billed.[2] Organizations that need limits beyond Tier 5 can submit a rate limit increase request form.[2]
Search and embeddings limits
The Search API has its own limit that applies to every tier: 50 query units per second with a burst of 50. Billing and rate limiting use different units, because a successful request with five queries is one billable request but consumes five rate-limit units.[2] Embeddings limits scale with tier: 85 QPS at Tier 0, 170 at Tiers 1 to 3 and 335 at Tiers 4 and 5, with contextualized embeddings limited by chunks per second at five times those figures.[2]
Credits, projects and roles
The API Portal at console.perplexity.ai is organized around projects. A project holds billing, API keys, members and usage monitoring.[3] The billing model is credit-based: credits are bought in advance, usage is charged by tokens processed and search queries made, and automatic top-up can be enabled. If credits run out, API keys are blocked until the balance is topped up.[3] Payments are managed through Stripe, and adding a payment method does not itself charge the card.[3] Members have the role Admin or Member: Admins can invite and remove members, manage billing and view usage data, while Members cannot access the Billing page or usage metrics.[3] The help center states that Automatic Top Up replenishes the balance when it drops below $2, and that refunds can be processed if reported within 14 days of purchase and the credits have not been used; used credits are non-refundable.[9] Enterprise Pro and Enterprise Max seats do not include API access or credits, and API Platform credits are billed separately from Enterprise seats.[10]
Virtualization & partitioning
Not applicable. The API Platform is hosted. The only partitioning concept is the project, which isolates keys, members and billing, and the organization-wide scope of Agent API rate limits.[3][2]
Cloud / BYOL
The platform can be purchased through AWS Marketplace as a SaaS listing. Credits bought there are denominated one to one with USD, are shared by the Orchestrator, Agent, Search and Embeddings APIs, and the listing is a one-month contract starting at USD 1,000 of credits.[4] Purchase alone does not provision anything: the buyer must complete a post-purchase registration, after which Perplexity creates the user and a new API project, makes that user its administrator, generates an initial key and applies the credits.[4] Marketplace credits are not added to an existing self-service project.[4] The page states that credits do not expire, because the one-month contract refers to the billing cycle, and that refunds are available within 14 days of purchase provided no credits have been used.[4] The listing is described as a separate product from Enterprise Pro, which does not include API access.[4]
Programs
- AWS Marketplace API credits: consolidated AWS billing, enterprise procurement and EDP eligibility.[4]
- API automatic top-up: optional replenishment of the balance when it falls below a threshold.[9]
- Custom rate limit request: a form for limits above Tier 5.[2]
Data handling
The documentation states that the Chat Completions API follows a zero data retention policy: Perplexity does not retain data sent through it and does not use customer data to train models, collecting only billable metrics such as tokens processed, model used, timestamps and the API key identifier.[5] The API Terms separately state that Perplexity will not use customer content to train generative AI models and that protected health information may not be processed without a Business Associate Agreement.[11] The security page lists a SOC 2 Type II report, a 2025 HIPAA gap assessment and a CAIQlite questionnaire.[5]
Out of scope
This article does not cover model-by-model third-party token rates, which Perplexity publishes on the Agent API Models page and which change often, the Analytics API, API key management mechanics, or the legal terms of the API Terms of Service, which are summarized in the terms article.