Social
🧵 jina-embeddings-v5-omni represents the next evolution of embedding models. Literally.
— Elastic (@elastic) August 17, 2026
The vision pipeline uses the Qwen3.5 vision encoder, which itself is based on SigLIP2.
Audio comes through the Qwen2.5-Omni audio encoder, built atop Whisper-large-3.
Text runs through… pic.twitter.com/dfKAPcDctg
When you create a recipe, you try a few tweaks and combinations before landing on the final version.
— Elastic (@elastic) August 13, 2026
Creating an embedding model is no different.
This process of experimentation is known as ablation.
For both audio and vision pipelines in jina-embeddings-v5-omni, multiple… pic.twitter.com/1zDS16vsom
How can a text embedding model map audio and image vectors?
— Elastic (@elastic) August 11, 2026
After being transformed via a projector (translator), a foreign modality vector must announce itself before entering an embedding model for an unrelated modality.
Think of HTML.
<p> These tags let the document know… pic.twitter.com/Qko9k2Dzhr
Retraining a model is an expensive process.
— Elastic (@elastic) August 7, 2026
LoRA adapters allow you to fine-tune an existing model without retraining.
For jina-embeddings-v5-omni, we use a LoRA adapter to tweak the final output to be optimized for 4 distinct tasks. pic.twitter.com/MSo2QCoVs0
VLM is a superset of CLIP, not a competitor. People pick between them like it's either/or. It isn't.
— Elastic (@elastic) July 28, 2026
Think of a standalone GPS unit (fast, cheap, narrow similarity scoring) versus a smartphone app (richer reasoning, at the cost of speed and compute).
Look inside jina-v5-omni:… pic.twitter.com/KK1g0UJsFn
🧵 Your search query gets rewritten before it ever matches a document.
— Elastic (@elastic) July 14, 2026
Tokenization, stop words, stemming, synonyms: 4 steps sit between what you type and what gets looked up.
Here's what each one does pic.twitter.com/cRfUvz0muV
Search feels simple until you start getting back irrelevant results.
— Elastic (@elastic) July 10, 2026
Know which of these 3 retrieval strategies to reach for a furniture store site:
- BM25 matches exact terms. Finds an ottoman from "Product ID 43926".
- Vector matches meaning. Figures out what "padded stool for… pic.twitter.com/FXNL2FgVto
A red cluster is a decision tree, not one magic command.
— Elastic (@elastic) July 9, 2026
Bookmark these common commands to help you diagnose cluster, node and shard health issues pic.twitter.com/qT9vpXYEYe