Elastic Introduces Jina v5 Omni Family: Two Models to Power Text, Image, Video, and Audio Search

New addition to the Jina v5 model family delivers flexible, cost-efficient AI search across media types without rebuilding existing systems

SAN FRANCISCO--(BUSINESS WIRE)--Elastic (NYSE: ESTC), the Search AI Company, today announced jina-embeddings-v5-omni, a new family of multimodal embedding models with the ability to represent text, images, video, and audio as vectors. Developers can now perform search, classification, clustering, and deduplication across different media types, giving users powerful new ways to understand and organize multimodal data.

Available in two sizes, small and nano, the new omni models share the exact same text embedding space as jina-embeddings-v5-text, so v5-text users can keep their existing index, swap in an omni model, and immediately index multimedia into the same vectors.

“Our goal with v5-omni is simple: make multimodal search as easy and scalable as text search already is,” said Ken Exner, chief product officer, Elastic. “By building on existing models and ensuring full compatibility, we’re giving teams a practical way to expand into images, audio, and video, without starting from scratch.”

Powered by a single universal language model that aligns all modalities, the v5-omni models also feature a modular design, so users can toggle text, image, and audio processing features as needed, offering flexibility and efficiency across a wide range of use cases. Key highlights include:

• Multimodal Search: Supports text, image, video, and audio inputs
• Best-in-class performance: For text, images, and audio recordings, v5-omni offers top performance for models of comparable size
• Flexible Performance: Offers adjustable embedding sizes, allowing users to balance accuracy, speed, and cost
• Efficiency at Scale: Optimized for lower storage and compute requirements through variable embedding sizes and quantization
• Global Capabilities: Built with strong multilingual capabilities across dozens of languages
• Seamless Upgrading: Maintains compatibility with existing v5 text embeddings, eliminating the need to re-index data

In independent evaluations, jina-embeddings-v5-omni provides frontier-class results across four modalities in one compact model:

• Audio: v5-omni models rank best in their size class for audio retrieval on the Massive Audio Embedding Benchmark (MAEB), outperforming systems many times its size.
• Image: On the Massive Image Embedding Benchmark (MIEB) and Visual Document Retrieval Benchmark (ViDoRe), v5-omni outperforms models up to 20x larger, including specialists that handle only a single modality. The v5-omni-small model is the top performing model in the 1B parameter range in visual similarity and image retrieval. In the multilingual sections of the MIEB, v5-omni models are state-of-the-art by a significant margin, outclassing models over double their size.
• Text: v5-omni retains the frontier-level performance of jina-embeddings-v5-text, leading its size class in text retrieval, and beating models in the 7B–14B range on the Massive Multilingual Text Embedding Benchmark (MMTEB).
• Video: On the Massive Multimodal Embedding Benchmark (MMEB-v2), v5-omni competes with video grounding models several times its size.

Availability

Both jina-embeddings-v5-omni-small and jina-embeddings-v5-omni-nano models are available on Elastic Inference Service, via the Jina API, and for local installation via download. Model weights are distributed freely for non-commercial license use. Contact Elastic sales for commercial use.

Additional Materials

• Elastic blog: One index, all media: Introducing jina-embeddings-v5-omni

About Elastic

Elastic (NYSE: ESTC), the Search AI Company, integrates its deep expertise in search technology with artificial intelligence to help everyone transform all of their data into answers, actions, and outcomes. Elastic's Search AI Platform — the foundation for its search, observability, and security solutions — is used by thousands of companies, including more than 50% of the Fortune 500. Learn more at elastic.co.

Elastic and associated marks are trademarks or registered trademarks of elasticsearch B.V. and its subsidiaries. All other company and product names may be trademarks of their respective owners.