jina-embeddings-v5-omni is here! Our first universal embedding model for text, images, audio, and video. Available in two sizes: small (1.57B, 1024-dim, 32K context) and nano (0.95B, 768-dim, 8K context). Both support Matryoshka truncation down to 32 dimensions. v5-omni is back-compatible: if you already use jina-embeddings-v5-text-small/nano, the existing text indexes... work with v5-omni out of the box. Without reindexing the text, just index your multimodal content with v5-omni and start searching images, audio, and video.show more

Pareto frontier of all open-weight omni embedding models (supporting text, image, audio, and video). jina-embeddings-v5-omni-small (1.57B) matches the average score of LCO-7B (8.93B) while using 5.7x fewer parameters. jina-embeddings-v5-omni-nano (0.95B) outperforms LanguageBind (1.14B) by +8.9 points.

Per-task performance across 13 task types. Gold stars mark tasks where jina-embeddings-v5-omni-small beats the best open-weight baseline (3-9x larger). Wins: image classification (68.55 vs 64.30), image clustering (84.57 vs 83.24), audio classification (55.89 vs 53.39). Main gaps: video retrieval (27.82 vs 58.73) and compositional/VQA (44.23 vs 53.40).

v5-omni keeps the v5-text backbone completely frozen and adds pretrained vision and audio encoders connected through small trainable projectors: - Vision: Qwen3.5 vision encoders with 2x2 spatial merge. We freeze everything except the final projection layer (fc_vision_2), which we replace with a randomly initialized layer mapping into the text backbone's hidden dimension. - Audio: Qwen2.5-Omni encoder. A single randomly initialized fc_audio layer projects the 1280-dimensional output into the text backbone. - Video: Handled as a sequence of visual frames, optionally preceded by an extracted audio segment.

Today v5-omni is available on Elastic Inference Service, HuggingFace and Jina API. Learn more about v5-omni from links below. 🤗: arXiv: blog:

You guys haven’t made small (<100M) models in quite some time. I’d love to see a small variant of your v5 text models.

this is great, how does it compare with cohere embed 4?

@MichelIvan92347 any benchmark available from official end against bms3?

id say multimodal retrieval only becomes boring enough to use when it fits existing infra. same vector space + matryoshka dims is the kind of unsexy detail that makes adoption much easier, im a big fan

Interesting check your dm

🤤

Awesome, I’ve been waiting for someone to do this

This is a big step toward truly universal embeddings. Native support for text, images, audio, and video in a single model opens up exciting possibilities for multimodal AI applications. 🔥

v5 omni covers text to video clean. matryoshka down to 32 dims keeps it efficient for rag stacks

tab complete spoiled me

Great, thank you.