Jina Reader iconJina Reader

Jina AI's Reader API: prefix any URL with r.jina.ai to get clean, LLM-ready Markdown; s.jina.ai adds web search. Apache-2.0 code

12K+ GitHub Stars
10M Free tokens per new key
20 RPM with no key at all

Overview

Jina Reader turns a web page into text an LLM can use: put https://r.jina.ai/ in front of any URL and it returns clean Markdown (or JSON, HTML or text). It renders pages in headless Chrome or fetches them directly, reads PDFs and Office files, can caption images that have no alt text, and lets you keep or drop parts of the page with CSS selectors. A second endpoint, s.jina.ai, searches the web and returns the top five results already read into text. You can also switch the conversion to ReaderLM-v2, Jina's 1.54B-parameter HTML-to-Markdown and JSON model. Basic use needs no API key; a key raises the rate limits and bills by output tokens. The service code is Apache-2.0 on GitHub and ships as a Docker image for self-hosting, without the hosted service's storage layer. Jina AI is now part of Elastic. Also from Jina AI: Jina AI (embeddings and rerankers).

The Verdict

Who Should Use Jina Reader?

Best For

  • Developers who want one URL in and Markdown out, with no SDK
  • RAG and agent tools that read single pages or search results on demand
  • Quick prototypes: works without signing up
  • Teams that want to self-host the same code under Apache 2.0

Not Ideal For

  • Crawling whole sites or clicking through pages (it reads one URL per call)
  • Sites behind CAPTCHAs, unless you bring your own proxy
  • Commercial use of ReaderLM-v2 weights without a license (CC BY-NC 4.0)

What's Great

  • Very simple: prefix a URL, read the response
  • Fine control through headers: selectors, engine, timeout, cache, cookies, proxy, token budget
  • Reads PDFs and Office files, not just HTML
  • Failed requests are not charged
  • Repeat requests within 5 minutes come back from cache almost at once

Watch Out For

  • Average latency listed at 7.9 seconds for Reader (2.5 for search)
  • No site crawler or browser agent for multi-page jobs
  • No public per-token price; you buy token top-ups
  • Apify's comparison found Firecrawl 4 to 5 times cheaper at about 100K pages a month (Apify sells a rival scraper)
  • OCR output mode uses 40 times more tokens

Pricing

View all features & details

Endpoints

  • r.jina.ai: read one URL to Markdown, HTML, text or JSON
  • s.jina.ai: web search, top five results read into text
  • X-Respond-With readerlm-v2 for model-based conversion and JSON-schema extraction

Request Controls

  • X-Target-Selector and X-Remove-Selector (CSS)
  • X-Engine: direct, browser (default) or Cloudflare browser rendering
  • X-Token-Budget, X-Timeout, X-No-Cache, X-Set-Cookie, X-Proxy-Url
  • X-With-Generated-Alt for image captions; X-Page for one PDF page

Self-Hosting

  • Docker image with headless Chrome, LibreOffice and CJK fonts
  • Optional S3-compatible bucket for caching
  • Some licensed assets (MaxMind data, a font) are downloaded separately
  • Open-source branch re-synced with the hosted code in April 2026

Real-World Usage

Adoption

  • Strong GitHub following for the Reader repo (see badge for count)
  • Last code push in May 2026

ReaderLM-v2

  • 1.54B-parameter model for HTML to Markdown or JSON
  • Weights on Hugging Face under CC BY-NC 4.0; commercial use needs a license

How It Compares

Feature Jina Reader Firecrawl Crawl4AI
Main job One URL to Markdown, plus search Scrape, crawl and extract Open-source crawling library
Whole-site crawl No Yes Yes
Use with no sign-up Yes (20 RPM) No Self-run
Free hosted allowance 10M tokens per new key 1,000 credits a month None (self-host)
Code license Apache 2.0 AGPL-3.0 Apache 2.0

Sources: Jina Reader · Firecrawl pricing · Firecrawl repo · Crawl4AI repo · Apify comparison

Is this your tool? Add a badge