jina-embeddings-v4: Universal Embeddings for Multimodal Multilingual Retrieval

By Michael Günther, Saba Sturua, Mohammad Kalim Akram, Isabelle Mohr, A. Ungureanu, Bo Wang, Sedigheh Eslami, Scott Martens, Maximilian Werk, Nan Wang and 1 more. 2025

Subfields: Information Retrieval & Search, Multimodal Learning, Representation & Self-Supervised Learning

Citations in the corpus

• SkMTEB: Slovak Massive Text Embedding Benchmark and Model Adaptation (2026, unknown): “( Sturua et al., 2024; Günther et al., 2025) Recent large-scale embedding families.”
• On The Role of Pretrained Language Models in General-Purpose Text Embeddings: A Survey (2025, unknown): “Text Pairs ( 𝑥, 𝑦 + / 𝑦 − ) Text Triples ( 𝑥, 𝑦 + , 𝑦 − ) Instructed Triples ( 𝐼 ◦ 𝑥, 𝑦 + , 𝑦 − ) mistral”
• VaRS-Doc: Interpretation-Aware Variant Representations via Latent Self-Probing for Visual Document Retrieval (2026, unknown): “Recent MLLM-based retrievers (Jiang et al. 2024, 2025; Günther et al. 2025; Nussbaum, Duderstadt, and Mul-yar 2024; Team 2025; Guan et al. 2026; Zeng et al. 2021, 2023, 2024; Qin et al. 2025), including ColPali (Faysse et al. 2025), ColQwen (Faysse et al. 2025; Bai et al. 2025), and VisRAG (Yu et…”
• LightSTAR: Efficient Visual Document Retrieval via Lightweight Selection with Vision-Adaptive Refinement (2026, unknown): “While these MLLM-based approaches achieve competitive retrieval accuracy on challenging benchmarks [9, 21, 35], a major obstacle remains: they require MLLM-level computation for every page during indexing and retrieval.”
• Visual Product Search Benchmark (2026, unknown): “It is a unified multimodal embedding model for universal retrieval across visual, textual, and multilingual content [12].”
• Unlocking Multimodal Document Intelligence: From Current Triumphs to Future Frontiers of Visual Document Retrieval (2026, unknown): “While recent benchmarks like Jina-VDR (Günther et al., 2025) and MIRACL-VISION (Osmulski et al., 2025) have made commendable strides in multi-lingual support, the majority of available resources remain predominantly English-centric and confined to general-domain web documents.”
• Nemotron ColEmbed V2: Top-Performing Late Interaction embedding models for Visual Document Retrieval (2026, unknown): “Notable implementations include ColPali [7] (PaliGemma-based [26]), ColQwen [7] (Qwen2-based [27]), Jina Embedding Models [28] (Qwen2.”
• FreshStack: Building Realistic Benchmarks for Evaluating Retrieval on Technical Documents (2025, unknown): “We have evaluated and included the following families of models: (i) Qwen3 embeddings (596M, 4B, and 8B) [93], (ii) Jina embeddings (V3 and V4) [70, 21], (iii) Stella (1.5B and 400M) [91], (iv) OpenAI text-embedding-3 (small and large) [53], (v) GTE-large-en-v1.”
• CORE-Bench: A Comprehensive Benchmark for Code Retrieval in the Era of Agentic Coding (2026, unknown): “(Günther et al., 2025; Akram et al., 2026), and F2LLM-v2 (Zhang et al., 2026) provide strong transfer baselines.”
• Evo-Retriever: LLM-Guided Curriculum Evolution with Viewpoint-Pathway Collaboration for Multimodal Document Retrieval (2026, unknown): “Unlike traditional document retrieval, CVDR faces unique challenges due to the heterogeneous structure of documents such as financial reports, contracts, and scientific papers [19, 21], where vital information is often dispersed across text, charts, and tables [9, 24].”