Jina.ai

NewsModelsAPIkeyboard_arrow_downReaderConvert any URL to Markdown for better grounding LLMs.EmbeddingsWorld-class multimodal multilingual embeddings.RerankerWorld-class reranker for maximizing search relevancy. MCP terminalCLIarticlellms.txtsmart_toyAgentsdata_objectSchemamenu_bookDocsLog inloginlanguageThemeroutineYour Search Foundation Supercharged.codeAPIcognitionModelsContactStart instantly—no credit card or registration needed!verified_userWe are SOC 2 Type 1 & 2 compliant with the American Institute of Certified Public Accountants (AICPA).open_in_newsmart_displayHow to get my API key?groupsOur CustomersloginkeyAPI Key & BillingReaderEmbeddingsRerankerchevron_leftchevron_righthomespeedRate Limitbug_reportRaise issuehelp_outlineFAQmenu_bookDocsarrow_drop_down Statuschevron_leftchevron_rightglobe_bookUse r.jina.ai to read a URL and fetch its contenttravel_exploreUse s.jina.ai to search the web and get SERPAdd mcp.jina.ai as your MCP server to access our API in LLMsParametersarrow_drop_downThe target URL to fetch content fromAdd API Key for Higher Rate Limit Enter your Jina API key to access a higher rate limit. For latest rate limit information, please refer to the table below.open_in_newLearn moreBrowser Engine (Quality/Speed) Choose the browser engine for fetching the webpage content. This affects the quality, speed, completeness, accessibility of the content.open_in_newLearn moreDefaultarrow_drop_downContent Format You can control the level of detail in the response to prevent over-filtering. The default pipeline is optimized for most websites and LLM input.Defaultarrow_drop_downJSON Response The response will be in JSON format, containing the URL, title, content, and timestamp (if available). In Search mode, it returns a list of five entries, each following the described JSON structure.Timeout (seconds) Maximum time to wait for page load. Increase for slow pages, decrease for simple static pages.Token Budget Limits the maximum number of tokens used for this request. Exceeding this limit will cause the request to fail.Use ReaderLM-v2 ExperimentalUses ReaderLM-v2 for HTML to Markdown conversion, to deliver high-quality results for websites with complex structures and contents. Costs 3x tokens!open_in_newLearn moreExtract Only (CSS Selector) Only extract content matching these CSS selectors. Example: article, .main-content, #post-bodybody .class #id Wait For (CSS Selector) Wait until these elements appear before extracting content. Useful for dynamically loaded content.body .class #id Exclude (CSS Selector) Remove these elements before extraction. Example: nav, footer, .sidebar, #adsheader .class #id Remove All Images Strip all images from the output. Reduces token usage when images are not needed.OpenAI Citation Format Format links for OpenAI's web browsing tool. Uses special citation markers compatible with GPT models.open_in_newLearn moreLinks Summary Section A "Buttons & Links" section will be created at the end. This helps the downstream LLMs or web agents navigating the page or take further actions.Nonearrow_drop_downImages Summary Section An "Images" section will be created at the end. This gives the downstream LLMs an overview of all visuals on the page, which may improve reasoning.Nonearrow_drop_downBrowser Viewport Size POSTSet browser window dimensions. Affects responsive layouts and content visibility.open_in_newLearn moreForward Cookie Our API server can forward your custom cookie settings when accessing the URL, which is useful for pages requiring extra authentication. Note that requests with cookies will not be cached.open_in_newLearn more= =; domain= Image Caption Captions all images at the specified URL, adding 'Image [idx]: [caption]' as an alt tag for those without one. This allows downstream LLMs to interact with the images in activities such as reasoning and summarizing.Use a Proxy Server Our API server can utilize your proxy to access URLs, which is helpful for pages accessible only through specific proxies.open_in_newLearn moreUse a Country-Specific Proxy Server Set country code for location-based proxy server. Use 'auto' for optimal selection or 'none' to disable.Bypass Cached Content Our API caches URL contents for a certain amount of time. Set it to true to ignore the cached result and fetch the content from the URL directly.Cache Tolerance (seconds) Accept cached content if younger than N seconds. Set to 0 for fresh content (same as Bypass Cache), or higher values to allow faster responses from cache.Page Ready Timing When to consider a page fully loaded. Later timings wait longer but capture more dynamic content.Defaultarrow_drop_downCustom User-Agent Override the browser User-Agent string. Useful for accessing sites that require specific browsers or block crawlers.Custom Referer Set the HTTP Referer header. Some sites check this to verify traffic comes from expected sources.Preserve Base64 Images Keep inline base64-encoded images in markdown output instead of converting them to external URLs.Do Not Cache or Track Prevent this request from being cached or logged on our servers. Use for sensitive URLs.Github Flavored Markdown Opt in/out features from GFM (Github Flavored Markdown).Enabledarrow_drop_downStream Mode Stream mode is beneficial for large target pages, allowing more time for the page to fully render. If standard mode results in incomplete content, consider using Stream mode.open_in_newLearn moreCustomize Browser Locale Control the browser locale to render the page. Lots of websites serve different content based on the locale.open_in_newLearn moreRespect robots.txt Check robots.txt rules before fetching. Specify which bot name to use for the check.Include iframe Content Extract content from embedded iframes. Enable for pages with content loaded in iframes.Include Shadow DOM Extract content from Shadow DOM components. Enable for pages using web components.Use Final URL as Base Resolve relative URLs using the final destination URL after redirects, instead of the original URL.Local PDF/HTML file POSTUse Reader on your local PDF and HTML file by uploading them. Only support pdf and html files. For HTML, please also specify a reference URL for better parsing related CSS/JS scripts.uploadRun JavaScript Before Extraction POSTExecute custom JS to modify the page before content extraction. Can be inline code or a URL to a script file.open_in_newLearn moreHeading Style Sets markdown heading format (passed to Turndown).Hash Stylearrow_drop_downHorizontal Rule Style Defines markdown horizontal rule format (passed to Turndown).Bullet Point Style Sets bullet list marker character (passed to Turndown).*arrow_drop_downEmphasis Style Defines markdown emphasis delimiter (passed to Turndown)._arrow_drop_downStrong Emphasis Style Sets markdown strong emphasis delimiter (passed to Turndown).**arrow_drop_downLink Style Determines markdown link format (passed to Turndown).Inlinearrow_drop_downEU Compliance ExperimentalAll infrastructure and data processing operations reside entirely within EU jurisdiction.uploadRequestGETBashLanguagearrow_drop_downwrap_textcurl "https://r.jina.ai/https://www.example.com" content_copysendGET RESPONSEkeyAPI keyvisibility_offcontent_copyAvailable tokens0 syncThis is your unique key. Store it securely!For Better SearchOur frontier models form the search foundation for high-quality enterprise search and RAG systems.ReaderConvert a URL to LLM-friendly input, by simply adding r.jina.ai in front.Get startedarrow_forwardEmbeddingsWorld-class multimodal multilingual embeddings.Get startedarrow_forwardRerankerWorld-class reranker for maximizing search relevancy.Get startedarrow_forwardOur PublicationsUnderstand how our frontier search models were trained from scratch, check out our latest publications. Meet our team at EMNLP, SIGIR, ICLR, NeurIPS, and ICML!schoolRead morearXiv February 17, 2026jina-embeddings-v5-text: Task-Targeted Embedding DistillationarXiv February 11, 2026Embedding Inversion via Conditional Masked Diffusion Language ModelsICLR 2026January 22, 2026Embedding Compression via Spherical CoordinatesarXiv December 29, 2025Vision Encoders in Vision-Language Models: A SurveyICLR 2026December 04, 2025Jina-VLM: Small Multilingual Vision Language ModelAAAI 2026October 01, 2025jina-reranker-v3: Last but Not Late Interaction for Document RerankingNeurIPS 2025August 31, 2025Efficient Code Embeddings from Code Generation ModelsEMNLP 2025June 24, 2025jina-embeddings-v4: Universal Embeddings for Multimodal Multilingual RetrievalICLR 2025March 04, 2025ReaderLM-v2: Small Language Model for HTML to Markdown and JSONACL 2025December 17, 2024AIR-Bench: Automated Heterogeneous Information Retrieval BenchmarkICLR 2025December 12, 2024jina-clip-v2: Multilingual Multimodal Embeddings for Text and ImagesECIR 2025September 18, 2024jina-embeddings-v3: Multilingual Embeddings With Task LoRASIGIR 2025September 07, 2024Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding ModelsEMNLP 2024August 30, 2024Jina-ColBERT-v2: A General-Purpose Multilingual Late Interaction RetrieverWWW 2025June 21, 2024Leveraging Passage Embeddings for Efficient Listwise Reranking with Large Language ModelsICML 2024May 30, 2024Jina CLIP: Your CLIP Model Is Also Your Text RetrieverarXiv February 26, 2024Multi-Task Contrastive Learning for 8192-Token Bilingual Text EmbeddingsarXiv October 30, 2023Jina Embeddings 2: 8192-Token General-Purpose Text Embeddings for Long DocumentsEMNLP 2023July 20, 2023Jina Embeddings: A Novel Set of High-Performance Sentence Embedding Models19 publications in total.Officeslocation_onSunnyvale, CA710 Lakeway Dr, Ste 200, Sunnyvale, CA 94085, USAlocation_onBerlin, GermanyPrinzessinnenstraße 19-20, 10969 Berlin, GermanySearch FoundationReaderEmbeddingsRerankerGet Jina API keyRate LimitAPI StatusCompanyAbout usContact salesNewsIntern programDownload Jina logoopen_in_newDownload Elastic logoopen_in_newTermsSecurityTerms & ConditionsPrivacyManage Cookies emailJina AI by Elastic © 2020-2026. --- The Future Starts HerehandshakeContact 2020Founded in Employees Publications Open Models Developers Empowered These photos include our former colleagues and interns—we appreciate every one of them. favoriteOur Mission: Search/accFounded in 2020 by Han Xiao, Jina AI is a leading search AI company. We provide Embeddings, Rerankers, Readers, and Small Language Models that help businesses and developers build powerful search applications—delivering highly relevant, grounded, multimodal, and multilingual results. On October 9, 2025, Jina AI was acquired by Elastic (NYSE: ESTC).star2025/05/05Read moreopen_in_newstar2024/10/13ondemand_videoVideo interviewRead moreopen_in_newstar2024/10/14Read moreopen_in_newstar2024/05/08Read moreopen_in_newstar2024/02/27ondemand_videoVideo interviewRead moreopen_in_new2021/04/21Read moreopen_in_newOur AwardsOur CustomersBusinesses of all sizes trust Jina AI's Search Foundation to power their tools and products—so can you.LLM ProvidersTechFinanceE-commerceMediaLLM ProvidersMajor LLM companies use Jina Reader to crawl and search the web for better training data.Tech CompaniesSoftware, cloud, and AI companies use our search tools to build their RAG and AI agent systems. Our reader, embeddings, and rerankers help them ship products faster.Finance & ConsultingBanks and consulting firms use our tools to clean data and get insights fast. We offer secure on-premise options so they can keep their data private.E-commerce & RetailOnline stores use Jina to power product search and recommendations. Our multilingual models help them sell globally.Media & PublishingNews and media companies use Jina to search through articles, videos, and archives. Our Reader and rerankers help them find exactly what they need.Other IndustriesCompanies in education, farming, real estate, and more use our tools to clean and search their data. We help them find what they need, faster.Our InvestorsOfficeslocation_onSunnyvale, CA710 Lakeway Dr, Ste 200, Sunnyvale, CA 94085, USAlocation_onBerlin, GermanyPrinzessinnenstraße 19-20, 10969 Berlin, GermanySearch FoundationReaderEmbeddingsRerankerGet Jina API keyRate LimitAPI StatusCompanyAbout usContact salesNewsIntern programDownload Jina logoopen_in_newDownload Elastic logoopen_in_newTermsSecurityTerms & ConditionsPrivacyManage Cookies emailJina AI by Elastic © 2020-2026. --- News Accelerate search AI, one token at a time.schoolAcademicrss_feedRSS folder_specialFeaturedjina-embeddings-v5-text: New SOTA Small Multilingual EmbeddingsTwo sub-1B multilingual embeddings with best-in-class performance, available on Elastic Inference Service, Llama.cpp and MLX.February 19, 2026 • 7 minutes readJina-VLM: Small Multilingual Vision Language ModelNew 2B vision language model achieves SOTA on multilingual VQA, no catastrophic forgetting on text-only tasks.December 04, 2025 • 7 minutes readJina Reranker v3: 0.6B Listwise Reranker for SOTA Multilingual RetrievalNew 0.6B-parameter listwise reranker that considers the query and all candidate documents in a single context window.October 03, 2025 • 7 minutes readschoolAcademic PublicationsarXiv February 17, 2026jina-embeddings-v5-text: Task-Targeted Embedding DistillationarXiv February 11, 2026Embedding Inversion via Conditional Masked Diffusion Language ModelsICLR 2026January 22, 2026Embedding Compression via Spherical CoordinatesarXiv December 29, 2025Vision Encoders in Vision-Language Models: A SurveyICLR 2026December 04, 2025Jina-VLM: Small Multilingual Vision Language ModelAAAI 2026October 01, 2025jina-reranker-v3: Last but Not Late Interaction for Document RerankingNeurIPS 2025August 31, 2025Efficient Code Embeddings from Code Generation ModelsEMNLP 2025June 24, 2025jina-embeddings-v4: Universal Embeddings for Multimodal Multilingual RetrievalICLR 2025March 04, 2025ReaderLM-v2: Small Language Model for HTML to Markdown and JSONACL 2025December 17, 2024AIR-Bench: Automated Heterogeneous Information Retrieval BenchmarkICLR 2025December 12, 2024jina-clip-v2: Multilingual Multimodal Embeddings for Text and ImagesECIR 2025September 18, 2024jina-embeddings-v3: Multilingual Embeddings With Task LoRASIGIR 2025September 07, 2024Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding ModelsEMNLP 2024August 30, 2024Jina-ColBERT-v2: A General-Purpose Multilingual Late Interaction RetrieverWWW 2025June 21, 2024Leveraging Passage Embeddings for Efficient Listwise Reranking with Large Language ModelsICML 2024May 30, 2024Jina CLIP: Your CLIP Model Is Also Your Text RetrieverarXiv February 26, 2024Multi-Task Contrastive Learning for 8192-Token Bilingual Text EmbeddingsarXiv October 30, 2023Jina Embeddings 2: 8192-Token General-Purpose Text Embeddings for Long DocumentsEMNLP 2023July 20, 2023Jina Embeddings: A Novel Set of High-Performance Sentence Embedding Models19 publications in total.folder_specialFeaturedschoolAcademicAllPress releaseTech blogEventOpinionchevron_leftchevron_rightMarch 11, 2026 • 7 minutes readBootstrapping Audio Embeddings from Multimodal LLMsTurn any multimodal LLM into a small audio embedding model that beats CLAP with 25x less data.March 06, 2026 • 6 minutes readIdentifying Embedding Models from Raw Numerical ValuesA tiny transformer that fingerprints embedding models by reading raw numerical digits. No feature engineering.February 19, 2026 • 7 minutes readjina-embeddings-v5-text: New SOTA Small Multilingual EmbeddingsTwo sub-1B multilingual embeddings with best-in-class performance, available on Elastic Inference Service, Llama.cpp and MLX.February 17, 2026jina-embeddings-v5-text: Task-Targeted Embedding DistillationText embedding models are widely used for semantic similarity tasks, including information retrieval, clustering, and classification. General-purpose models are typically trained with single- or multi-stage processes using contrastive loss functions. We introduce a novel training regimen that combines model distillation techniques with task-specific contrastive loss to produce compact, high-performance embedding models. Our findings suggest that this approach is more effective for training small models than purely contrastive or distillation-based training paradigms alone. Benchmark scores for the resulting models, jina-embeddings-v5-text-small and jina-embeddings-v5-text-nano, exceed or match the state-of-the-art for models of similar size. jina-embeddings-v5-text models additionally support long texts (up to 32k tokens) in many languages, and generate embeddings that remain robust under truncation and binary quantization. Model weights are publicly available, hopefully inspiring further advances in embedding model development.arXiv February 11, 2026Embedding Inversion via Conditional Masked Diffusion Language ModelsWe frame embedding inversion as conditional masked diffusion, recovering all tokens in parallel through iterative denoising rather than sequential autoregressive generation. A masked diffusion language model is conditioned on the target embedding via adaptive layer normalization, requiring only 8 forward passes through a 78M parameter model with no access to the target encoder. On 32-token sequences across three embedding models, the method achieves 81.3% token accuracy and 0.87 cosine similarity.arXiv January 22, 2026Embedding Compression via Spherical CoordinatesWe present a compression method for unit-norm embeddings that achieves 1.5x compression, 25% better than the best prior lossless method. The method exploits that spherical coordinates of high-dimensional unit vectors concentrate around pi/2, causing IEEE 754 exponents to collapse to a single value and high-order mantissa bits to become predictable, enabling entropy coding of both. Reconstruction error is below 1e-7, under float32 machine epsilon. Evaluation across 26 configurations spanning text, image, and multi-vector embeddings confirms consistent improvement. The method requires no training.ICLR 2026December 29, 2025Vision Encoders in Vision-Language Models: A SurveyVision encoders have remained comparatively small while language models scaled from billions to hundreds of billions of parameters. This survey analyzes vision encoders across 70+ vision-language models from 2023–2025 and finds that training methodology matters more than encoder size: improvements in loss functions, data curation, and feature objectives yield larger gains than scaling by an order of magnitude. Native resolution handling improves document understanding, and multi-encoder fusion captures complementary features no single encoder provides. We organize encoders into contrastive, self-supervised, and LLM-aligned families, providing a taxonomy and practical selection guidance for encoder design and deployment.arXiv December 04, 2025 • 7 minutes readJina-VLM: Small Multilingual Vision Language ModelNew 2B vision language model achieves SOTA on multilingual VQA, no catastrophic forgetting on text-only tasks.December 04, 2025Jina-VLM: Small Multilingual Vision Language ModelWe present jina-vlm, a 2.4B parameter vision-language model that achieves state-of-the-art multilingual visual question answering among open 2B-scale VLMs. The model couples a SigLIP2 vision encoder with a Qwen3 language backbone through an attention-pooling connector that enables token-efficient processing of arbitrary-resolution images. Across standard VQA benchmarks and multilingual evaluations, jina-vlm achieves leading results while preserving competitive text-only performance. Model weights and code are publicly released.ICLR 2026October 03, 2025 • 7 minutes readJina Reranker v3: 0.6B Listwise Reranker for SOTA Multilingual RetrievalNew 0.6B-parameter listwise reranker that considers the query and all candidate documents in a single context window.123…12Search by titlesearchFilter by productarrow_drop_downFilter by authorarrow_drop_downFilter by modelarrow_drop_downOfficeslocation_onSunnyvale, CA710 Lakeway Dr, Ste 200, Sunnyvale, CA 94085, USAlocation_onBerlin, GermanyPrinzessinnenstraße 19-20, 10969 Berlin, GermanySearch FoundationReaderEmbeddingsRerankerGet Jina API keyRate LimitAPI StatusCompanyAbout usContact salesNewsIntern programDownload Jina logoopen_in_newDownload Elastic logoopen_in_newTermsSecurityTerms & ConditionsPrivacyManage Cookies emailJina AI by Elastic © 2020-2026. --- Our Search Foundation ModelsWe've been moving the needle in search models since day one. Take a look at our model evolution below—hover or click to discover each milestone.Browse catalogHelp me decide2023Q22023Q42024Q12024Q22024Q32024Q42025Q12025Q22025Q32025Q42026Q1calendar_month2023-06-17jina-embedding-b-en-v1notes512background_dot_small768data_array110M🇺🇸calendar_month2023-10-28jina-embeddings-v2-base-ennotes8Kbackground_dot_small768data_array137M🇺🇸calendar_month2024-01-09jina-embeddings-v2-base-zhnotes8Kbackground_dot_small768data_array161M🇺🇸 🇨🇳calendar_month2024-01-15jina-embeddings-v2-base-denotes8Kbackground_dot_small768data_array161M🇺🇸 🇩🇪calendar_month2024-02-05jina-embeddings-v2-base-codenotes8Kbackground_dot_small768data_array137M🇺🇸calendar_month2024-02-14jina-embeddings-v2-base-esnotes8Kbackground_dot_small768data_array161M🇺🇸 🇪🇸calendar_month2024-02-17jina-colbert-v1-engrid_4x4Multi-vectornotes8Kdata_array137M🇺🇸calendar_month2024-02-29jina-reranker-v1-base-ennotes8Kdata_array137M🇺🇸calendar_month2024-04-18jina-reranker-v1-tiny-ennotes8Kdata_array33M🇺🇸calendar_month2024-04-18jina-reranker-v1-turbo-ennotes8Kdata_array37.8M🇺🇸calendar_month2024-06-05jina-clip-v1mmsMultimodalnotes8Kbackground_dot_small768data_array223M🇺🇸copyrightcalendar_month2024-06-25jina-reranker-v2-base-multilingualnotes1Kdata_array278M🌍copyrightcalendar_month2024-08-11reader-lm-0.5bnotes256Kdata_array494M🌍copyrightcalendar_month2024-08-11reader-lm-1.5bnotes256Kdata_array1.54B🌍copyrightcalendar_month2024-08-31jina-colbert-v2grid_4x4Multi-vectornotes8Kdata_array560M🌍copyrightcalendar_month2024-09-18jina-embeddings-v3notes8Kbackground_dot_small1024data_array570M🌍copyrightcalendar_month2024-11-05jina-clip-v2mmsMultimodalnotes8Kbackground_dot_small1024data_array865M🌍copyrightcalendar_month2025-01-16ReaderLM-v2notes512Kdata_array1.54B🌍copyrightcalendar_month2025-04-08jina-reranker-m0mmsMultimodalnotes10Kdata_array2.4B🌍calendar_month2025-06-24jina-embeddings-v4mmsMultimodalnotes32Kbackground_dot_small2048data_array3.8B🌍copyrightcalendar_month2025-09-01jina-code-embeddings-0.5bnotes32Kbackground_dot_small896data_array494M🌍copyrightcalendar_month2025-09-01jina-code-embeddings-1.5bnotes32Kbackground_dot_small1536data_array1.5B🌍copyrightcalendar_month2025-10-01jina-reranker-v3notes131Kdata_array597M🌍copyrightcalendar_month2025-12-04jina-vlmnotes32Kdata_array2.4B🌍copyrightcalendar_month2026-02-18jina-embeddings-v5-text-nanonotes8Kbackground_dot_small768data_array239M🌍copyrightcalendar_month2026-02-18jina-embeddings-v5-text-smallnotes32Kbackground_dot_small1024data_array677M🌍 Press enter or space to select a node. You can then use the arrow keys to move the node around. You can then use the arrow keys to move the node around, press delete to remove it and press escape to cancel. Press enter or space to select an edge. You can then press delete to remove it or press escape to cancel. searchsortDatearrow_drop_downjina-embeddings-v5-text-nanojina-embeddings-v5-text-smalljina-vlmjina-reranker-v3jina-code-embeddings-0.5bjina-code-embeddings-1.5bjina-embeddings-v4jina-reranker-m0ReaderLM-v2jina-clip-v2jina-embeddings-v3jina-colbert-v2reader-lm-0.5breader-lm-1.5bjina-reranker-v2-base-multilingualjina-clip-v1jina-reranker-v1-tiny-enjina-reranker-v1-turbo-enjina-reranker-v1-base-enjina-colbert-v1-enjina-embeddings-v2-base-esjina-embeddings-v2-base-codejina-embeddings-v2-base-dejina-embeddings-v2-base-zhjina-embeddings-v2-base-enjina-embedding-b-en-v1Select a model from the list to view detailsOfficeslocation_onSunnyvale, CA710 Lakeway Dr, Ste 200, Sunnyvale, CA 94085, USAlocation_onBerlin, GermanyPrinzessinnenstraße 19-20, 10969 Berlin, GermanySearch FoundationReaderEmbeddingsRerankerGet Jina API keyRate LimitAPI StatusCompanyAbout usContact salesNewsIntern programDownload Jina logoopen_in_newDownload Elastic logoopen_in_newTermsSecurityTerms & ConditionsPrivacyManage Cookies emailJina AI by Elastic © 2020-2026.