Newsroom

Multilingual models often face a "language gap," where similar phrases in different languages don't align. We show how contrastive learning can bridge this gap, enhancing cross-language performance.

Part 2 of our exploration of Late Chunking, a deep dive into why it is the best method for chunk embeddings and improving search/RAG performance.

We collected some tips to help you migrate from Jina Embeddings v2 to v3.

jina-embeddings-v3 is a frontier multilingual text embedding model with 570M parameters and 8192 token-length, outperforming the latest proprietary embeddings from OpenAI and Cohere on MTEB.

Reader-LM-0.5B and Reader-LM-1.5B are two novel small language models inspired by Jina Reader, designed to convert raw, noisy HTML from the open web into clean markdown.

Jina ColBERT v2 supports 89 languages with superior retrieval performance, user-controlled output dimensions, and 8192 token-length.

You can't just use a CLIP model to retrieve text and images and sort the results by score. Why? Because of the modality gap. What is it, and where does it come from?

Chunking long documents while preserving contextual information is challenging. We introduce the "Late Chunking" that leverages long-context embedding models to generate contextual chunk embeddings for better retrieval applications.

What does it mean for LLMs when the web has been strip-mined clean, content providers have locked their doors, and there’s barely a trickle of new data to scrape?

We had a blast at ICML 2024 in Vienna, and we want to share with you everything we said, saw, and learned.

When using embedding models for zero-shot classification, rephrasing the class label to "This is seriously about 'LABEL'" gives higher accuracy vs. using LABEL alone. But how, and why?

A lot of LLMs can't figure out that 9.11 is actually smaller than 9.9. Can our embedding and reranker models do any better?