Jina AI builds search foundation models for RAG and retrieval pipelines: multilingual text and multimodal embeddings (jina-embeddings-v3, jina-embeddings-v4 at 3.8B), ColBERT-style late-interaction retrievers (Jina-ColBERT-v2), listwise rerankers (jina-reranker-v3, a 0.6B Qwen3-based "last but not late interaction" model), and ReaderLM for HTML-to-markdown conversion. Acquired by Elastic in 2025.
Jina AI released jina-ocr-v1, a 3.4B mixture-of-experts document parser with speculative decoding that turns pages into Markdown at 2.57 pages per second.
Jina AI's 278M parameter cross-encoder reranks documents across 100+ languages, plus function calling and code search, with 6x faster throughput than v1.
Jina AI's compact 33M-parameter embedding model packs 8192-token context into a tiny footprint, using bidirectional ALiBi instead of positional embeddings.