Skip to content

Jina Reader

Put r.jina.ai/ in front of any URL and get the page back as clean Markdown, with a search endpoint alongside.

Reviewed by Andrew Miller · Updated

Visit Jina Reader Pay per use

Our rating

4.0/5

Jina Reader turns one URL into clean Markdown for fractions of a cent, which makes it the right default for agents that read a page at a time and the wrong tool for anyone who needs to crawl a site.

What it is

Jina Reader is a hosted API from Jina AI, now part of Elastic, that converts a web page or PDF into Markdown for language models. You prefix a URL with https://r.jina.ai/ and it strips navigation, headers, footers and ads. A sister endpoint, s.jina.ai, runs a web search and returns the top five results with their content.

Behaviour is set through request headers: CSS selectors to keep or remove, waiting for an element, keeping images and links, image captions, proxies, cookies, caching and a token budget. JSON output and streaming are available, and the ReaderLM-v2 model (1.54 billion parameters) can be switched on for harder HTML-to-Markdown or schema-based extraction.

Without a key, Reader allows 20 requests a minute; a key raises that to 500 (5,000 on premium) and comes with free tokens from a balance shared with Jina's embedding and reranking models. Since 10 August 2026 Elastic sells commercial and on-premise licences priced annually. The server code is on GitHub under Apache 2.0.

Our review

Jina Reader wins on friction. There is no SDK and no account needed to try it: one prefixed URL returns usable Markdown, which is why it turns up in so many agent prototypes. Billing is by output tokens; Jina's January 2025 deployment guide listed $50 for 1 billion tokens and $500 for 11 billion ($0.045 to $0.05 per million), and a 2026 comparison by Parallel reports the same rates, so an ordinary page costs well under a cent, and failed requests are not charged.

It is a page reader, not a crawler. There is no site-wide crawl or sitemap mapping as in Firecrawl, search stops at five results and costs a fixed 10,000 tokens or more per query, and Jina's own page lists average Reader latency of 7.9 seconds with the default browser engine. The direct engine returns simple pages in a few hundred milliseconds, so latency is partly a configuration choice spread across a long list of X- headers rather than a typed client.

Two data defaults deserve attention. Jina says it never trains on API inputs or outputs, but repeat requests for the same URL within five minutes are served from cache and requests are logged unless you send the DNT header; requests carrying cookies are never cached. An EU residency option is labelled experimental, and deleting an account is a manual process that takes a few business days.

The open-source label oversells what you can run yourself. The Apache 2.0 repository's last push was in May 2026, the hosted service is the product, and ReaderLM-v2's weights are CC BY-NC 4.0, which Jina itself says is not an open-source licence. Since August 2026, commercial self-hosting runs through an annual Elastic licence, a very different purchase from a $0.05-per-million-token API.

Where it earns its place

  • A developer building a research agent on top of Claude or GPT

    Let the agent read any link a user pastes

    Wrap fetches as GET https://r.jina.ai/<url> with an API key for 500 requests a minute. A 2,000-token page costs about $0.0001; slow JavaScript-heavy pages are the main failure mode, so set a timeout.

  • A data engineer refreshing a RAG index from 50,000 documentation pages

    Convert known URLs to Markdown each week without running a headless browser fleet

    At about 2,000 output tokens a page, a weekly run is roughly 100 million tokens, about $5. It needs the URL list up front, since Reader cannot discover pages, and X-No-Cache to avoid five-minute stale copies.

  • A product manager at a legal-tech start-up handling client documents

    Decide whether page content can pass through a third-party reader

    Jina does not train on requests, but logs and caches them unless DNT is set, and EU residency is experimental. For privileged material the on-premise Elastic licence is the route, priced annually by sales.

Before you commit

Setup
Prefix any URL with https://r.jina.ai/ in a browser or curl; add a Bearer key for higher limits.
Integrations
Plain HTTP with headers, JSON and streaming output, s.jina.ai search, ReaderLM-v2 and jina-vlm for harder pages; also on AWS, Azure and GCP marketplaces.
Data and privacy
No training on inputs or outputs. Responses cached for five minutes and requests logged unless DNT is sent; experimental EU residency.
Cost at scale
One million pages a month at about 2,000 output tokens each is roughly 2 billion tokens, about $100. OCR via jina-ocr-v1 costs 40 times more tokens.
Limits
20 RPM without a key, 500 with one, 5,000 on premium; search 100 RPM on standard keys; concurrency around 4,000 requests.

What's good

  • One-line use with no SDK or sign-up
  • Fractions of a cent per typical page
  • Reads PDFs as well as HTML
  • Shares a token balance with Jina embeddings and rerankers
  • No training on API inputs or outputs

What's not

  • No whole-site crawling
  • 7.9-second average latency on the default engine
  • Search capped at five results, 10,000+ tokens each
  • Requests logged and cached unless you send DNT
  • ReaderLM-v2 weights are non-commercial; self-hosting needs an Elastic licence

Best for

Developers who need an agent or pipeline to read individual web pages as Markdown at minimal cost.

Alternatives

Jina Reader: common questions

Is Jina Reader worth it?
Jina Reader turns one URL into clean Markdown for fractions of a cent, which makes it the right default for agents that read a page at a time and the wrong tool for anyone who needs to crawl a site. We rate it 4.0 out of 5.
How much does Jina Reader cost?
Free without a key at 20 requests/minute; keys include free tokens, then usage is billed by output tokens from a shared Jina balance (Jina's 2025 guide listed $50 for 1 billion tokens; current rates sit behind the dashboard). Search costs at least 10,000 tokens per request. Commercial and on-premise licences sold annually by Elastic.
What are the downsides of Jina Reader?
No whole-site crawling; 7.9-second average latency on the default engine; search capped at five results, 10,000+ tokens each; requests logged and cached unless you send DNT; readerLM-v2 weights are non-commercial; self-hosting needs an Elastic licence.
Who is Jina Reader best for?
Developers who need an agent or pipeline to read individual web pages as Markdown at minimal cost.
How much does Jina Reader cost for a team or in production?
One million pages a month at about 2,000 output tokens each is roughly 2 billion tokens, about $100. OCR via jina-ocr-v1 costs 40 times more tokens.
Is Jina Reader safe for company data?
No training on inputs or outputs. Responses cached for five minutes and requests logged unless DNT is sent; experimental EU residency.
What are the best Jina Reader alternatives?
The closest alternatives we have reviewed are Apify.

Answers checked Oct 23, 2026.

Make Jina Reader? Show this rating on your siteJina Reader rated 4.0 out of 5 on ReviewRamble