jina-embeddings-v4
Jina's 3.8B multimodal embedding model unifying text and image retrieval, Apache 2.0.
About
jina-embeddings-v4 is a 3.8B-parameter multimodal embedding model from Jina AI, introduced with the arXiv paper in June 2025. It unifies text and image representations in one architecture supporting both single-vector and multi-vector (late interaction, ColBERT-style) embeddings, with task-specific LoRA adapters for retrieval, text matching, and code search.
Jina reports state-of-the-art results on cross-modal retrieval with particular strength on visually rich content — tables, charts, diagrams, mixed-media documents — measured on their new Jina-VDR benchmark. The model is Apache 2.0, which is unusually permissive for a model of this size, though at 3.8B parameters it wants a GPU for comfortable inference.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- SOTA on Jina-VDR and ViDoRe visual-document retrieval (Jina-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (metrics tier, metrics-v1). Evidence: Jina AI release with peer-reviewed paper (arXiv 2506.18902); 300K+ monthly HF downloads; Apache 2.0, unusually permissive for a 3.8B model; Jina-reported SOTA on Jina-VDR and ViDoRe; GPU recommended for inference. License: Apache 2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Jina AI release with peer-reviewed paper (arXiv 2506.18902); 300K+ monthly HF downloads
- · Apache 2.0, unusually permissive for a 3.8B model
- · Jina-reported SOTA on Jina-VDR and ViDoRe; GPU recommended for inference