Jina AI Free API Key: 10M Free Tokens for Embeddings, Reranker & Reader (2026)

Short answer: yes, Jina AI gives you a free API key, no sign-up form or card needed. The key is generated automatically on Jina's website and comes with 10 million free tokens that work across all Jina APIs: embeddings, reranking, Reader (turn any URL into clean text) and Search. Keys never expire. On the free key, the Embedding and Reranker APIs allow 100 requests and 100,000 tokens per minute. Jina's Reader also works with no key at all, at 20 requests per minute.

Checked against Jina's Embeddings, Reader and Reranker pages on 2026-10-17. Earlier we could not find the 10M figure on an official page; Jina's Reader page now states it directly.

What you get for free

• An auto-generated API key. Visit any Jina API page and a key is shown, ready to copy.
• 10M free tokens per new key. The Reader page says every new API key comes with 10M free tokens. The Embeddings page adds that the free tokens are usable across any of Jina's models.
• No expiry on the key. Jina says its API keys do not have an expiration date.
• No charge for failures. Tokens are not deducted for failed requests.
• No training on your data. Jina's FAQ says it never uses API requests, inputs or outputs to train its models.
• Keyless Reader. Put https://r.jina.ai/ in front of a URL and you get LLM-friendly text back, free for basic use at 20 requests per minute.

What you do not get

• No monthly refill. Jina's pages describe the free tokens as coming with a new key. They do not describe any monthly reset, so plan as if the 10M is a one-time balance.
• No self-hosting rights. Jina says token purchases cover use of the hosted APIs only. Running Jina model weights on your own servers is a separate licensing question (see below).
• Lower rate limits than paid keys. Free keys get 100 requests per minute on embeddings and reranking; paid keys get 500, and premium keys 5,000.

Rate limits by API

Jina tracks limits in requests per minute (RPM) and tokens per minute (TPM), by IP address or API key, whichever limit is reached first. These are the figures from Jina's rate-limit table:

On top of the per-key limits there is an IP-based limit of 10,000 requests per 60 seconds to prevent abuse. Jina's general FAQ quotes 100 RPM for free keys, while the Reader row of its own table says 500 RPM with a free key; the table is the more specific source for Reader.

How far 10 million tokens goes

Because every API draws from the same balance, it helps to know what each call costs. Some rough arithmetic using Jina's own counting rules:

Your real numbers depend on your text, but the pattern is clear: embeddings are the cheapest use of free tokens, and Search is the most expensive. If you mainly want to embed a few thousand documents for a prototype, 10M tokens is plenty. If you run Search in an agent loop, it can be used up in an afternoon.

How to get and use the free key, step by step

1. Open jina.ai/embeddings (or the Reader or Reranker page). An API key is generated automatically and shown on the page.
2. Copy the key and store it as an environment variable, for example JINA_API_KEY. Anyone with the key can spend its tokens, so treat it like a password.
3. Call the API with the key as a Bearer token (example below).
4. To check your balance, top up, or revoke a key, log in to Jina's API Key Management dashboard. Revoking disables the key immediately and its remaining balance can no longer be used.
5. When the free tokens run out, buy more for the same key in the Buy tokens tab. Payments go through Stripe; depending on where you are you may be charged in USD, EUR or another currency, and taxes may apply.

curl https://api.jina.ai/v1/embeddings \
  -H "Authorization: Bearer $JINA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "jina-embeddings-v5-text-small", "input": ["How do I reset my password?"]}'

For Reader, no code is needed at all: open https://r.jina.ai/https://example.com in a browser or with curl and you get the page as clean text. Add your key in an Authorization header to get the higher rate limit. If you request the same URL again within 5 minutes, Reader returns the cached content.

Models you can use

Licensing and commercial use

There are two separate questions here. Using Jina's hosted API is covered by your tokens: Jina says token purchases cover use of the hosted Jina APIs. Running Jina's model weights yourself is different. Many Jina models are released under CC BY-NC 4.0, which allows free non-commercial use but requires a commercial license for production use. Elastic sells that license as Jina On-Prem, available since August 10, 2026, and also runs Jina models inside Elastic Cloud through the Elastic Inference Service. A few models built on other bases have their own licenses; for example, Jina notes that the Qwen Research License does not permit commercial use.

Some third-party sites say Jina's free tokens are for non-commercial use only. We did not find that condition on Jina's own pages. If you plan to use the free key in a commercial product, read Jina's current terms first.

Jina AI and Elastic

On October 9, 2025, Elastic announced that it had completed its acquisition of Jina AI, bringing Jina's multilingual and multimodal embedding models and rerankers into Elastic's search platform. The jina.ai APIs, free keys and token top-ups are still available, and Jina's pages now point enterprise questions, on-premises licenses and account security issues to Elastic Sales and Elastic Support.

How Jina compares with other free embedding APIs

Voyage gives far more free tokens if you only need embeddings and reranking. Jina's advantage is that one key also covers Reader and Search, which is handy for retrieval-augmented generation (RAG) prototypes: fetch a page with Reader, embed it, then rerank results, all on one balance. For more options, see our guide to free AI API keys.

Practical tips

• Keep one key per project. Keys do not expire, so it is easy to lose track. Separate keys make it simple to revoke one without affecting others.
• Batch your embedding requests. Sending many texts per request keeps you under 100 requests per minute while staying within 100,000 tokens per minute.
• Use Reader without a key for light scraping. If you only need a few pages a minute, the keyless 20 RPM limit saves your token balance.
• Watch Search costs. Each s.jina.ai request costs at least 10,000 tokens, so cache results and avoid calling it in tight loops.
• Do not commit the key to a public repo. Anyone who finds it can use your tokens until you revoke it.

Common mistakes

• Expecting a monthly refill. The 10M tokens come with a new key; Jina does not describe a monthly reset.
• Confusing API use with self-hosting. Downloading CC BY-NC weights for a commercial product needs a commercial license, even if you used the API for free.
• Revoking a key with paid tokens on it. Revocation cannot be undone. Premium keys can transfer their remaining paid balance to another key first; do that before revoking.