Perplexity launches Context API to extract web pages at $1 per 1,000 URLs

Perplexity launched the Context API on November 17, 2026, a developer endpoint that takes a list of web addresses and returns the content of each page as clean Markdown, raw HTML, a list of links or a set of passages selected for a query. The company prices it at $1 per 1,000 fetched URLs, or $0.001 per page, with no token billing, and it charges only for URLs that come back successfully. Pages that fail, time out or hit a bot challenge are free.
The endpoint lives at api.perplexity.ai/v1/context and is aimed at AI agents that need to read a specific page rather than search for one: a coding agent checking the parameters of an API, a support agent reading a refund policy, or a research agent working through a full report instead of a snippet. Each request can carry up to 10 URLs on the entry usage tier and up to 25 on the highest tiers, and the results come back in the same order as the input.
Perplexity describes the Context API as drawing on its own web index when a recent copy exists and fetching the page live when it does not. By default the service accepts a stored copy up to two days old, and developers can force a live read or restrict the call to cached copies with two request parameters. In an internal benchmark on 1,000 URLs from real agent traffic, Perplexity reports that 92.3% of its outputs were usable, with 87.9% fact recall and 80.3% content precision.
The launch arrived on the same day as two other Perplexity releases, the Orchestrator API for routing requests across frontier and open-weight models and a Multiagent effort level in Perplexity Computer for Max subscribers. Context API is the one with a published per-unit price, and at $1 per 1,000 pages it matches the rate of Perplexity's Fast Search tier and of its Monitor API.
··········
WHAT THE PERPLEXITY CONTEXT API RETURNS FOR EACH URL.
A Context API request is a POST call with a JSON body containing a urls array and, optionally, a formats array. Four output formats are available. Markdown is the default and returns the page text with navigation, cookie notices and other boilerplate removed. HTML returns the raw page markup for teams that prefer to parse it themselves. Links returns the absolute URLs found on the page, deduplicated and in page order. Highlights returns the passages most relevant to a query the developer supplies, up to a per-page token budget.
The highlights format selects passages from the page and keeps their original wording, which Perplexity presents as a way to give a model only the relevant parts of a long page without paraphrase. The query can run from 1 to 1,024 characters, and the per-page budget ranges from 1 to 4,096 tokens, with a default of 1,024. Passing an empty formats list returns metadata only.
To build the Markdown version, Perplexity says the service splits each page into blocks and uses a model to score every block as main content or boilerplate, keeping only the content blocks. Every successful result also carries metadata extracted from the page, including the title, description, language, author, published date, content type and Open Graph tags, plus two fields that record whether the copy came from the cache or from a live fetch and when it was fetched.
........
Format | What it returns | Typical use |
markdown (default) | Page text with boilerplate removed | Feeding a full page to a model |
html | Raw page HTML | Custom parsing and scraping logic |
links | Absolute URLs, deduplicated, in page order | Crawling and link discovery |
highlights | Query-relevant passages, original wording | Long pages with a token budget |
[] (empty) | Metadata only | Checking titles, dates and authors |
........
The API reference lists ten status types for individual results. Beyond ok, they include not_modified, invalid_url, blocked, bot_challenge, crawl_failed, crawl_timeout, no_text, no_content and internal_error. A failed result carries only the URL and its status, so an agent can see exactly which pages need a retry. The documentation does not describe how the service handles robots.txt rules, JavaScript-heavy pages or paywalls, and developers who depend on those cases will need to test them directly.
··········
CONTEXT API FRESHNESS, CACHING AND RATE LIMITS.
Freshness is controlled per request. The max_age_seconds parameter defaults to 172,800 seconds, which is two days, and accepts values from 0 to 63,072,000. Setting it to 0 forces a live fetch, which Perplexity recommends for prices, scores and breaking news, while a long value suits documentation and reference pages that rarely change. The cache_only flag prevents any live request and returns no_content when no stored copy exists. A third option, if_modified_since, lets a developer check a single URL against a date and receive not_modified instead of the full page when nothing has changed.
URLs in the same request are fetched in parallel, each with its own timeout. The default per-URL timeout is 30 seconds and it can be set between 1 and 110 seconds. A slow page therefore does not hold up the rest of the batch, and an agent can resend only the URLs that timed out. Request bodies larger than 512 KiB are rejected.
Rate limits follow Perplexity's API usage tiers. Each request counts as one unit regardless of how many URLs it contains, so batching pages into a single call stretches the limit further.
........
Usage tier | Requests per second | URLs per request | Max URLs per second |
Tier 0 | 1 | 10 | 10 |
Tiers 1-3 | 3 | 15 | 45 |
Tiers 4-5 | 5 | 25 | 125 |
........
Perplexity's quickstart states that raised limits can reach 50 URLs per request for accounts that request them. Calls above the rate limit return HTTP 429 with a Retry-After header.
··········
PERPLEXITY CONTEXT API PRICING AND HOW BILLING WORKS.
The pricing page lists the Context API at $1.00 per 1,000 fetched URLs, which is $0.001 per URL, with no token billing. A URL is billed when its status is ok or not_modified. Every other status, from blocked to crawl_timeout, costs nothing. The rate applies per URL, and the pricing page lists no surcharge by output format, so a request for Markdown, links and highlights on the same page is billed as one fetched URL.
Each successful response includes a usage object with the number of billed URLs and an estimated cost in USD at list price. Perplexity notes that this figure is an estimate and not the amount charged, a distinction for teams with negotiated or volume pricing.
The Context API fills a gap in Perplexity's developer catalog between search and monitoring. The table below shows how it sits next to the company's other web-data products, using prices from the official pricing page on November 18, 2026.
........
Perplexity product | Billing unit | Price |
Context API | Per 1,000 fetched URLs | $1.00 |
Search API, Standard | Per 1,000 requests | $5.00 |
Search API, Fast Search | Per 1,000 requests | $1.00 |
Monitor API, Search run | Per 1,000 succeeded runs | $1.00 |
Monitor API, Page run | Per 1,000 counted URLs | $1.00 |
Agent API web_search tool | Per 1,000 invocations | $2.50 |
Agent API fetch_url tool | Per 1,000 invocations | $0.50 |
........
The fetch_url tool in the Agent API costs less per call, at $0.0005, but it runs inside an Agent API request, where model tokens are billed on top. The Context API is a standalone endpoint with no model charges, which makes its cost predictable for pipelines that only need page content. Readers following Perplexity's developer releases can compare it with the Monitor API launched a day earlier at the same $1 rate and with the recent Agent API ultra effort and price changes.
··········
CONTEXT API COST COMPARED WITH FIRECRAWL AND EXA.
Perplexity's benchmark names Firecrawl, Exa, Parallel, Keenable and Context.dev as the services it tested against. Firecrawl and Exa publish clear per-page prices, so they make the most direct comparison. The figures below come from each company's official pricing page on November 18, 2026, and they compare list prices, not output quality, which Perplexity's benchmark does not let readers verify for the other providers.
Firecrawl sells monthly plans with credits. A basic scrape costs 1 credit per page, and the JSON, Question and Highlight formats add 4 credits per page. Pages that return a 403 or 404 error still cost 1 credit. Exa bills its Contents endpoint at $1 per 1,000 pages for each content type, so a page requested with both text and highlights counts twice.
To show the difference in practice, consider a research pipeline that reads 300,000 pages a month, with every page fetched successfully. The calculation uses monthly billing for Firecrawl and the cheapest plan or plan-plus-top-up combination that covers the volume.
........
Workload (300,000 pages a month) | Perplexity Context API | Exa Contents | Firecrawl |
Plain page text or Markdown | $300 | $300 | $399 (Growth plan, 500,000 credits) |
Text plus query-based highlights | $300 | $600 | $1,249 (Scale plan plus top-ups) |
Failed pages | Not billed | Not stated | 403 and 404 billed at 1 credit |
........
For plain Markdown, Perplexity and Exa land at the same $300 a month, and Firecrawl costs $399 because the volume falls between its Standard plan and its Growth plan. Covering the same 300,000 pages on Firecrawl Standard would take $99 plus 200,000 top-up credits at $5 per 2,000, or $599, so the larger plan is cheaper. The gap widens once highlights enter the request. Exa counts text and highlights as two content types and doubles to $600, while Firecrawl's Highlight format raises each page to 5 credits, or 1.5 million credits a month, which costs $749 for the Scale plan's first million plus $500 in top-ups at $5 per 5,000 credits. On Perplexity's published pricing, the highlights request stays at $300.
Firecrawl's plans include crawl and map features, concurrency allowances of up to 100 requests and unused-credit rules that a per-URL comparison does not capture, and Exa bundles its contents pricing with a search product that has its own rates. The table measures the cost of reading pages only.
··········
WHAT CHANGES FOR DEVELOPERS BUILDING AGENTS ON PERPLEXITY.
Developers who already hold a Perplexity API key can call the Context API immediately. The launch post does not mention any regional or plan restrictions, and the endpoint authenticates with the same Bearer key used by the Search, Agent and Monitor APIs. The quickstart shows the Python SDK pattern below, condensed from Perplexity's documentation.
from perplexity import Perplexity
client = Perplexity() # reads PERPLEXITY_API_KEY from the environment
response = client.context.create(urls=["https://example.com"])
for result in response.results:
if result.status.type == "ok":
print(result.markdown)The most common pattern pairs the Context API with a search call. An agent that runs one Standard Search API request and then reads the top five results through the Context API pays $0.005 for the search and $0.005 for the five pages, or $0.01 per task. At 10,000 tasks a month that comes to $100. Switching the search step to Fast Search, at $1 per 1,000 requests, brings the same workload to $60, because the page reads become the larger share of the bill.
Teams moving from self-hosted scraping should check three behaviors before switching. The first is freshness, since the default two-day cache can return an older copy of a page that changes often unless the request sets max_age_seconds to 0. The second is coverage of sites that block bots, which return bot_challenge and are not billed but also deliver no content. The third is batch size, since Tier 0 accounts are limited to one request per second with 10 URLs each, or 10 pages per second, which may not suit a large backfill without a tier upgrade.
For consumers rather than developers, nothing changes in the Perplexity app or in the Free, Pro and Max subscriptions. Readers comparing those plans can use our guide to Perplexity pricing and subscriptions.
··········
PERPLEXITY ORCHESTRATOR AND MULTIAGENT, RELEASED THE SAME DAY.
Perplexity paired the Context API with the Orchestrator API, which gives developers one endpoint for models from OpenAI, Anthropic, Google and xAI as well as open-weight models such as GLM, Kimi and DeepSeek, served by Perplexity or its inference partners. Orchestrator accepts the OpenAI Chat Completions, OpenAI Responses and Anthropic Messages formats, retries failed requests on another deployment of the same model, and adds spend breakdowns, model allowlists and spend caps in the API Console. Perplexity says there is no platform fee on top of credits but does not list per-model prices in the announcement. Two features, Routes and Auto effort, are in preview.
On the consumer side, Perplexity added a Multiagent effort level above Ultra in Perplexity Computer. A lead agent running on Claude Fable 5.2 plans the work and coordinates up to 16 agents, with a default team of Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Carbon that share findings and check each other's output. Multiagent is available to all Perplexity Max subscribers on the web, and Perplexity has not published credit costs for it.
··········
PERPLEXITY CONTEXT API: CONFIRMED FACTS AND OPEN QUESTIONS.
The verified parts of the launch are the price, the parameters and the limits. The Context API costs $0.001 per successfully fetched URL with no token charges, accepts 10 to 25 URLs per request depending on the usage tier, returns Markdown, HTML, links or query-based highlights, and lets developers choose between cached copies up to two days old and live fetches. Failed pages are free, which separates it from Firecrawl's treatment of 403 and 404 responses.
Three points remain unverified. Perplexity's benchmark figures of 92.3% usable output, 87.9% fact recall and 80.3% content precision come from its own test with a judge model, and the post does not publish the scores of the competing services, so the claimed lead cannot be checked. The documentation is silent on robots.txt, JavaScript rendering and paywalled pages, three areas that decide whether a page extraction service works for a given site list. Perplexity has also not said whether volume discounts apply to the Context API beyond the general invitation to contact its API team.
On list prices, the Context API is the cheapest of the three services compared here for workloads that need highlights, and it ties with Exa for plain page text. For teams already using Perplexity's Search or Monitor APIs, it completes a pipeline where search, page reading and change tracking all bill at predictable per-unit rates from a single account.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org




