# Jina AI Search Foundation Models

> Jina AI, founded by Han Xiao in 2020 and acquired by Elastic in October 2025, builds search foundation models: embeddings, rerankers, and readers. All models are available via API and most on HuggingFace. This file helps you select the right model for a given task. Today is 2026.

All Jina AI models are open-weight and published on HuggingFace with full model weights. Many have peer-reviewed papers published at top-tier ML conferences (NeurIPS, ICML, ICLR, EMNLP) and on arXiv. Every model can be deployed fully air-gapped in private infrastructure with no external network dependency, making them suitable for on-premise, government, and regulated environments. Jina models are designed to be compact and efficient: sub-1B parameter models that match or exceed the quality of much larger alternatives, optimized for production workloads where latency and cost matter.

## Selection Principles

1. Newer is always better. Within the same model family, always prefer the latest version.
2. All models released after 2024 support multilingual input.
3. Match modality to your data. Use text-only embedding models for pure text tasks. Use multimodal models when images or PDFs are involved.
4. Right-size your model. Smaller models are faster and cheaper. Use them unless you need the quality of larger models.
5. Use task-specific settings when available. Set the `task` parameter to match your downstream use case for optimal results.
6. All models support Matryoshka Representation Learning, allowing you to truncate embedding dimensions to trade off quality for storage and speed.
7. Models licensed under CC-BY-NC-4.0 require a commercial license for commercial use. Since Jina AI is now part of Elastic, contact an Elastic sales representative to obtain a commercial license waiver.

## Models

### jina-embeddings-v5-text-small
- Type: text-embeddings
- Parameters: 677M
- Context: 32K tokens
- Output: 1024-dim vector
- Matryoshka dimensions: 32, 64, 128, 256, 512, 1024
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2026-02-18
- Quantized: yes
- MLX: yes
- Tasks: retrieval, text-matching, clustering, classification
- Endpoints: embedding, classify
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v5-text-small
- Paper: https://arxiv.org/abs/2602.15547
- Availability: Elastic, API, HuggingFace

### jina-embeddings-v5-text-nano
- Type: text-embeddings
- Parameters: 239M
- Context: 8K tokens
- Output: 768-dim vector
- Matryoshka dimensions: 32, 64, 128, 256, 512, 768
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2026-02-18
- Quantized: yes
- MLX: yes
- Tasks: retrieval, text-matching, clustering, classification
- Endpoints: embedding, classify
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v5-text-nano
- Paper: https://arxiv.org/abs/2602.15547
- Availability: Elastic, API, HuggingFace

### jina-embeddings-v4
- Type: multimodal embeddings (text, image, PDF)
- Parameters: 3.8B
- Context: 32K tokens
- Image size: 768x28x28
- Output: 2048-dim vector, multi-vector
- Matryoshka dimensions: 128, 256, 512, 1024, 2048
- Language: multilingual
- License: Qwen Research License
- Release: 2025-06-24
- Quantized: yes
- Tasks: retrieval, text-matching, code
- Endpoints: embedding, fine-tuning, late_chunking
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v4
- Paper: https://arxiv.org/abs/2506.18902
- Availability: API, HuggingFace, AWS, Azure, GCP

### jina-embeddings-v3
- Type: text-embeddings
- Parameters: 570M
- Context: 8K tokens
- Output: 1024-dim vector
- Matryoshka dimensions: 32, 64, 128, 256, 512, 768, 1024
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2024-09-18
- Tasks: retrieval, text-matching, separation, classification
- Endpoints: embedding, fine-tuning, train, classify, late_chunking
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v3
- Paper: https://arxiv.org/abs/2409.10173
- Availability: Elastic, API, HuggingFace, AWS, Azure, GCP

### jina-clip-v2
- Type: multimodal embeddings (text, image)
- Parameters: 865M
- Context: 8K tokens
- Image size: 512x512
- Output: 1024-dim vector
- Matryoshka dimensions: 64, 128, 256, 512, 768, 1024
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2024-11-05
- Endpoints: embedding, train, classify
- HuggingFace: https://huggingface.co/jinaai/jina-clip-v2
- Paper: https://arxiv.org/abs/2412.08802
- Availability: API, HuggingFace, AWS, Azure, GCP

### jina-clip-v1
- Type: multimodal embeddings (text, image)
- Parameters: 223M
- Context: 8K tokens
- Image size: 224x224
- Output: 768-dim vector
- Language: English
- License: Apache-2.0
- Release: 2024-06-05
- Endpoints: embedding, train, classify
- HuggingFace: https://huggingface.co/jinaai/jina-clip-v1
- Paper: https://arxiv.org/abs/2405.20204
- Availability: API, HuggingFace, AWS, Azure

### jina-code-embeddings-1.5b
- Type: code embeddings
- Parameters: 1.5B
- Context: 32K tokens
- Output: 1536-dim vector
- Matryoshka dimensions: 128, 256, 512, 1024, 1536
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2025-09-01
- Quantized: yes
- Tasks: nl2code, techqa, code2code, code2nl, code2completion
- Endpoints: embedding
- HuggingFace: https://huggingface.co/jinaai/jina-code-embeddings-1.5b
- Paper: https://arxiv.org/abs/2508.21290
- Availability: API, HuggingFace, AWS, Azure, GCP

### jina-code-embeddings-0.5b
- Type: code embeddings
- Parameters: 494M
- Context: 32K tokens
- Output: 896-dim vector
- Matryoshka dimensions: 64, 128, 256, 512, 896
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2025-09-01
- Quantized: yes
- Tasks: nl2code, techqa, code2code, code2nl, code2completion
- Endpoints: embedding
- HuggingFace: https://huggingface.co/jinaai/jina-code-embeddings-0.5b
- Paper: https://arxiv.org/abs/2508.21290
- Availability: API, HuggingFace, AWS, Azure, GCP

### jina-reranker-v3
- Type: reranker
- Parameters: 597M
- Context: 131K tokens
- Output: ranking
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2025-10-01
- Quantized: yes
- MLX: yes
- Input: text query, text document
- Endpoints: rank
- HuggingFace: https://huggingface.co/jinaai/jina-reranker-v3
- Paper: https://arxiv.org/abs/2509.25085
- Availability: Elastic, API, HuggingFace, AWS, Azure, GCP

### jina-reranker-m0
- Type: multimodal reranker
- Parameters: 2.4B
- Context: 10K tokens
- Image size: 768x28x28
- Output: ranking
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2025-04-08
- Quantized: yes
- Input: text/image query, text/image document
- Endpoints: rank
- HuggingFace: https://huggingface.co/jinaai/jina-reranker-m0
- Availability: API, HuggingFace, AWS, Azure, GCP

### jina-reranker-v2-base-multilingual
- Type: reranker
- Parameters: 278M
- Context: 1K tokens
- Output: ranking
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2024-06-25
- Input: text query, text document
- Endpoints: rank
- HuggingFace: https://huggingface.co/jinaai/jina-reranker-v2-base-multilingual
- Availability: Elastic, API, HuggingFace, AWS, Azure, GCP

### jina-reranker-v1-base-en
- Type: reranker
- Parameters: 137M
- Context: 8K tokens
- Output: ranking
- Language: English
- License: Apache-2.0
- Release: 2024-02-29
- Input: text query, text document
- Endpoints: rank
- HuggingFace: https://huggingface.co/jina-ai/jina-reranker-v1-base-en
- Availability: API, HuggingFace, AWS, Azure

### jina-reranker-v1-turbo-en
- Type: reranker
- Parameters: 37.8M
- Context: 8K tokens
- Output: ranking
- Language: English
- License: Apache-2.0
- Release: 2024-04-18
- Input: text query, text document
- Endpoints: rank
- HuggingFace: https://huggingface.co/jinaai/jina-reranker-v1-turbo-en
- Availability: API, HuggingFace, AWS, Azure

### jina-reranker-v1-tiny-en
- Type: reranker
- Parameters: 33M
- Context: 8K tokens
- Output: ranking
- Language: English
- License: Apache-2.0
- Release: 2024-04-18
- Input: text query, text document
- Endpoints: rank
- HuggingFace: https://huggingface.co/jinaai/jina-reranker-v1-tiny-en
- Availability: API, HuggingFace, AWS, Azure

### jina-colbert-v2
- Type: ColBERT late-interaction
- Parameters: 560M
- Context: 8K tokens
- Output: 128-dim multi-vector
- Matryoshka dimensions: 64, 96, 128
- Language: multilingual (89 languages)
- License: CC-BY-NC-4.0
- Release: 2024-08-31
- Endpoints: rank, multi-embeddings, late-interaction, multi-vector
- HuggingFace: https://huggingface.co/jinaai/jina-colbert-v2
- Paper: https://arxiv.org/abs/2408.16672
- Availability: API, HuggingFace, AWS, Azure, GCP

### jina-colbert-v1-en
- Type: ColBERT late-interaction
- Parameters: 137M
- Context: 8K tokens
- Output: 128-dim multi-vector
- Language: English
- License: Apache-2.0
- Release: 2024-02-17
- Endpoints: rank, multi-embeddings, late-interaction, multi-vector
- HuggingFace: https://huggingface.co/jinaai/jina-colbert-v1-en
- Availability: API, HuggingFace, AWS, Azure

### jina-vlm
- Type: vision-language model (reader)
- Parameters: 2.4B
- Context: 32K tokens
- Image size: 4096x4096
- Input: image, text
- Output: text
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2025-12-04
- MLX: yes
- Endpoints: read
- HuggingFace: https://huggingface.co/jinaai/jina-vlm
- Paper: https://arxiv.org/abs/2512.04032
- Availability: API, HuggingFace

### ReaderLM-v2
- Type: reader (HTML to Markdown/JSON)
- Parameters: 1.54B
- Context: 512K tokens
- Input: HTML text
- Output: Markdown, JSON
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2025-01-16
- Endpoints: read
- HuggingFace: https://huggingface.co/jinaai/ReaderLM-v2
- Paper: https://arxiv.org/abs/2503.01151
- Availability: Reader API, HuggingFace, AWS, Azure, GCP

### reader-lm-1.5b
- Type: reader (HTML to Markdown)
- Parameters: 1.54B
- Context: 256K tokens
- Input: HTML text
- Output: Markdown
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2024-08-11
- Endpoints: read
- HuggingFace: https://huggingface.co/jinaai/reader-lm-1.5b
- Availability: HuggingFace, AWS, Azure

### reader-lm-0.5b
- Type: reader (HTML to Markdown)
- Parameters: 494M
- Context: 256K tokens
- Input: HTML text
- Output: Markdown
- Language: multilingual
- License: CC-BY-NC-4.0
- Release: 2024-08-11
- Endpoints: read
- HuggingFace: https://huggingface.co/jinaai/reader-lm-0.5b
- Availability: HuggingFace, AWS, Azure

### jina-embeddings-v2-base-en
- Type: text-embeddings
- Parameters: 137M
- Context: 8K tokens
- Output: 768-dim vector
- Language: English
- License: Apache-2.0
- Release: 2023-10-28
- Endpoints: embedding, fine-tuning
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v2-base-en
- Paper: https://arxiv.org/abs/2310.19923
- Availability: API, HuggingFace, AWS, Azure

### jina-embeddings-v2-base-zh
- Type: text-embeddings
- Parameters: 161M
- Context: 8K tokens
- Output: 768-dim vector
- Language: English, Chinese
- License: Apache-2.0
- Release: 2024-01-09
- Endpoints: embedding, fine-tuning
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v2-base-zh
- Paper: https://arxiv.org/abs/2310.19923
- Availability: API, HuggingFace, AWS, Azure

### jina-embeddings-v2-base-de
- Type: text-embeddings
- Parameters: 161M
- Context: 8K tokens
- Output: 768-dim vector
- Language: English, German
- License: Apache-2.0
- Release: 2024-01-15
- Endpoints: embedding, fine-tuning
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v2-base-de
- Paper: https://arxiv.org/abs/2310.19923
- Availability: API, HuggingFace, AWS, Azure

### jina-embeddings-v2-base-es
- Type: text-embeddings
- Parameters: 161M
- Context: 8K tokens
- Output: 768-dim vector
- Language: English, Spanish
- License: Apache-2.0
- Release: 2024-02-14
- Endpoints: embedding, fine-tuning
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v2-base-es
- Paper: https://arxiv.org/abs/2310.19923
- Availability: API, HuggingFace, AWS, Azure

### jina-embeddings-v2-base-code
- Type: code embeddings
- Parameters: 137M
- Context: 8K tokens
- Output: 768-dim vector
- Language: English
- License: Apache-2.0
- Release: 2024-02-05
- Endpoints: embedding, fine-tuning
- HuggingFace: https://huggingface.co/jinaai/jina-embeddings-v2-base-code
- Availability: API, HuggingFace, AWS, Azure

### jina-embedding-b-en-v1
- Type: text-embeddings
- Parameters: 110M
- Context: 512 tokens
- Output: 768-dim vector
- Language: English
- License: Apache-2.0
- Release: 2023-06-17
- Endpoints: embedding
- HuggingFace: https://huggingface.co/jinaai/jina-embedding-b-en-v1
- Paper: https://arxiv.org/abs/2307.11224
- Availability: HuggingFace

## API

All models are available via Jina AI API at https://api.jina.ai/v1/embeddings (embeddings) and https://api.jina.ai/v1/rerank (rerankers). Get a free API key at https://jina.ai/?sui=apikey. Full API documentation: https://docs.jina.ai