Jina AI API Pricing Calculator
Estimate your monthly token spend across Jina Embeddings, Reranker, Reader, DeepSearch, and Classifier APIs. See how Jina compares to OpenAI and Cohere on cost and benchmark performance.
Estimate Your Monthly Cost
Enter your expected usage across all Jina APIs
v4 is the flagship multimodal model; v3 is the best value for text-only RAG
1M tokens ≈ ~750K words. A 500-word doc ≈ 650 tokens
v3 is the fastest with SOTA BEIR performance; ColBERT uses late interaction for deeper matching
Typically 10–30% of your embedding volume in a retrieve-then-rerank pipeline
Each page extraction uses ~2,000–8,000 tokens; average ~4,000 tokens per page
ReaderLM-v2 gives better results on complex JS-heavy pages but costs 3× more tokens
Each query averages ~40,000 tokens (4 reasoning steps). Complex queries can use up to 300K tokens
Zero-shot and few-shot classification billed at same rate as embeddings
Tier unlocks higher rate limits — Free: 100 RPM, Paid: 500 RPM, Premium: 5,000 RPM
Cost Estimate
Monthly Jina AI API breakdown
Enter your API usage volumes across Embeddings, Reranker, Reader, and DeepSearch, then click Calculate to see your estimated monthly cost.
Jina AI Pricing by API & Model
Token rates and rate limits for all Jina APIs as of June 2026. All APIs share a single token pool from one API key. Verify current rates at jina.ai before committing budget.
| API / Model | Rate (per 1M tokens) | Context | Free Allowance | Notes |
|---|---|---|---|---|
| jina-embeddings-v4 | ~$0.02 | 32K | 10M tokens per new key | Multimodal, 3.8B params, SOTA MTEB |
| jina-embeddings-v3 | ~$0.02 | 8K | Best value; 65.5 MTEB, 570M params | |
| jina-reranker-v3 | ~$0.02 | 131K | Listwise, 0.6B params, BEIR SOTA | |
| jina-reranker-m0 | ~$0.02 | — | Multimodal visual document reranking | |
| jina-clip-v2 | ~$0.02 | 8K | Image + text, 865M params, 89 langs | |
| Reader API (standard) | ~$0.02 | 512K | Free unauthenticated | ~4K tokens per page avg; free without key |
| Reader API (ReaderLM-v2) | ~$0.06 (3×) | 512K | Free unauthenticated | Higher quality HTML→Markdown conversion |
| DeepSearch (jina-deepsearch-v1) | ~$0.02 | — | Shared token pool | Avg 40K tokens/query; max ~300K |
| Classifier API | ~$0.02 | 8K | Shared token pool | Zero-shot (256 classes) or few-shot (16) |
| OpenAI text-embedding-3-small (comparison) | $0.02 | 8K | None | Same price; lower MTEB than Jina v3/v4 |
| Cohere embed-english-v3 (comparison) | $0.10 | 512 | Trial | 5× more expensive than Jina |
| Voyage AI voyage-3-large (comparison) | $0.18 | 32K | None | 9× more expensive; highest MTEB score |
Jina AI Pricing FAQ
Everything developers and engineering teams need to know about Jina AI API costs, token budgeting, and rate limits in 2026.
Jina AI uses a token-based top-up model. Every new API key includes 10 million free tokens shared across all APIs. Beyond the free tier, tokens are purchased in bundles starting at 1B tokens. The effective rate is approximately $0.02 per 1M tokens for embeddings and reranking — making Jina one of the most cost-effective search foundation providers available in 2026. Jina updated its pricing model in May 2025, so always check your billing dashboard for current rates.
Yes. A single Jina API key gives access to all Search Foundation products — Embeddings, Reranker, Reader, DeepSearch, and Classifier — and token usage is pooled across all services. You do not need separate keys or separate budgets for each API. This makes it easy to build multi-step RAG pipelines without managing multiple billing accounts.
The Reader API (r.jina.ai) is free for unauthenticated use subject to IP-based rate limits. When you supply an API key, tokens are charged based on the length of the extracted markdown content returned. Using the ReaderLM-v2 engine for high-quality HTML-to-Markdown conversion costs 3× the standard token rate — worth it for complex JS-heavy pages, but unnecessary for clean article sites.
Jina AI rate limits are tiered by key type. Free: 100 RPM, 100K TPM, 2 concurrent requests. Paid: 500 RPM, 2M TPM, 50 concurrent requests. Premium: 5,000 RPM, 50M TPM, 500 concurrent requests. There is also an IP-based cap of 10,000 requests per 60 seconds across all APIs. For sustained production workloads above 500 RPM, you need a Premium key.
Jina embeddings-v3 and v4 are priced at approximately $0.02 per 1M tokens — identical to OpenAI’s text-embedding-3-small. However, Jina’s models score higher on MTEB benchmarks (65.5+ for v3) and support much longer context windows (8K–32K tokens vs OpenAI’s 8K). Against Cohere at $0.10/M and Voyage AI at $0.18/M, Jina is 5–9× cheaper at comparable benchmark quality.
Jina DeepSearch is an iterative search-read-reason API that loops until it finds an accurate answer to a query. It is billed against the same shared token pool at the same ~$0.02/M rate. A single DeepSearch query can consume between 10,000 and 300,000 tokens depending on complexity (averaging 4 reasoning steps and around 40,000 tokens). It is best suited for high-value research queries, not high-volume automated pipelines.
No. Jina AI does not deduct tokens for failed API requests. You are only charged for successfully processed tokens. This applies across all APIs including Embeddings, Reranker, Reader, and DeepSearch — which means retry logic is safe to implement without unexpected billing surprises.
Yes. Jina AI models are available on AWS SageMaker and Microsoft Azure Marketplace, allowing you to deploy models within your own cloud account and bill through your existing cloud provider. This is particularly useful for enterprise teams with data residency requirements, compliance needs, or existing cloud spending commitments. Google Cloud availability is also planned.