Model Catalog

All 28 models supported by jina-on-prem. Auto-generated from models/catalog.json - re-run python3 scripts/gen_catalog_md.py to refresh.

License note: Models tagged CC-BY-NC-4.0 need a commercial license for production use. Contact Elastic sales.

Embeddings

Model	Prebuilt	Params	VRAM	Context	Output	Modality	License
jina-embeddings-v5-omni-small	cpu / gpu	1.74B	~8GB	32K	1024 (matryoshka: 32-1024)	multimodal	CC-BY-NC-4.0
jina-embeddings-v5-omni-nano	cpu / gpu	1.04B	~5GB	8K	768 (matryoshka: 32-768)	multimodal	CC-BY-NC-4.0
jina-embeddings-v5-text-small	cpu / gpu	677M	~3GB	32K	1024 (matryoshka: 32-1024)	text	CC-BY-NC-4.0
jina-embeddings-v5-text-nano	cpu / gpu	239M	~2GB	8K	768 (matryoshka: 32-768)	text	CC-BY-NC-4.0
jina-code-embeddings-1.5b	cpu / gpu	1.5B	~4GB	32K	1536 (matryoshka: 128-1536)	code	CC-BY-NC-4.0
jina-code-embeddings-0.5b	cpu / gpu	494M	~2GB	32K	896 (matryoshka: 64-896)	code	CC-BY-NC-4.0
jina-embeddings-v4	cpu / gpu	3.8B	~10GB	32K	2048 (matryoshka: 128-2048)	multimodal	Qwen Research License
jina-clip-v2	cpu / gpu	865M	~4GB	8K	1024 (matryoshka: 64-1024)	multimodal	CC-BY-NC-4.0
jina-embeddings-v3	cpu / gpu	570M	~3GB	8K	1024 (matryoshka: 32-1024)	text	CC-BY-NC-4.0
jina-clip-v1	cpu / gpu	223M	~1GB	8K	768	multimodal	Apache-2.0
jina-embeddings-v2-base-es	cpu / gpu	161M	~1GB	8K	768	text	Apache-2.0
jina-embeddings-v2-base-code	cpu / gpu	137M	~1GB	8K	768	code	Apache-2.0
jina-embeddings-v2-base-de	cpu / gpu	161M	~1GB	8K	768	text	Apache-2.0
jina-embeddings-v2-base-zh	cpu / gpu	161M	~1GB	8K	768	text	Apache-2.0
jina-embeddings-v2-base-en	cpu / gpu	137M	~1GB	8K	768	text	Apache-2.0
jina-embedding-b-en-v1	cpu / gpu	110M	~1GB	512	768	text	Apache-2.0

Rerankers

Model	Prebuilt	Params	VRAM	Context	Output	Modality	License
jina-reranker-v3	cpu / gpu	597M	~3GB	128K	-	text	CC-BY-NC-4.0
jina-reranker-m0	cpu / gpu	2.4B	~6GB	10K	-	multimodal	CC-BY-NC-4.0
jina-reranker-v2-base-multilingual	cpu / gpu	278M	~1GB	1K	-	text	CC-BY-NC-4.0
jina-reranker-v1-turbo-en	cpu / gpu	37.8M	~1GB	8K	-	text	Apache-2.0
jina-reranker-v1-tiny-en	cpu / gpu	33M	~1GB	8K	-	text	Apache-2.0
jina-reranker-v1-base-en	cpu / gpu	137M	~1GB	8K	-	text	Apache-2.0

ColBERT

Model	Prebuilt	Params	VRAM	Context	Output	Modality	License
jina-colbert-v2	cpu / gpu	560M	~3GB	8K	128 (matryoshka: 64-128)	text	CC-BY-NC-4.0
jina-colbert-v1-en	cpu / gpu	137M	~1GB	8K	128	text	Apache-2.0

Readers

Model	Prebuilt	Params	VRAM	Context	Output	Modality	License
ReaderLM-v2	cpu / gpu	1.54B	~4GB	512K	-	text	CC-BY-NC-4.0
reader-lm-1.5b	cpu / gpu	1.54B	~4GB	256K	-	text	CC-BY-NC-4.0
reader-lm-0.5b	cpu / gpu	494M	~2GB	256K	-	text	CC-BY-NC-4.0

Vision-Language

Model	Prebuilt	Params	VRAM	Context	Output	Modality	License
jina-vlm	cpu / gpu	2.4B	~6GB	32K	-	multimodal	CC-BY-NC-4.0

Picking a model

Quick rules of thumb:

• First-time test / latency-critical: jina-embeddings-v5-text-nano (239M, ~2GB, CPU-friendly).
• Multilingual production embeddings: jina-embeddings-v5-text-small or jina-embeddings-v4.
• Multimodal (text + image): jina-embeddings-v5-omni-small or jina-clip-v2.
• Code search: jina-code-embeddings-1.5b (or 0.5b for smaller deploys).
• Reranking after retrieval: jina-reranker-v3 (best quality) or jina-reranker-v2-base-multilingual (faster).
• HTML/document cleanup: ReaderLM-v2 (largest context) or reader-lm-0.5b (lightweight).

See API Reference for the request shapes each model expects.