Embeddings & retrieval

Embedding and reranker models that power semantic search and RAG, plus the vector databases, serving engines and frameworks that use them, locally or in the cloud.

Models are ranked by MTEB(Multilingual, v2), then MTEB(eng, v2) for models without a result on the first. Tools inherit the tier of the best model they offer. How we rank.

Best overallSharrier-oss-v1-27b74.3% on MTEB(Multilingual, v2)
Best valueNo qualifying option yet
Best free startSOpenRouterFree tier · paid plans · via Qwen3-Embedding-8B
Best local (≤24 GB GPU)SKaLM-Embedding-Gemma3-12B-2511~24 GB VRAM · 72.3% on MTEB(Multilingual, v2)

Tier: how close to the best result on the task's main public benchmark
Outlined: a tool inherits the tier of the best model it offers
No public result yet
Cloud · Local: runs on your own hardware

Models

The engines themselves. Use them through an API or inside the apps on the right.

Cloud

–

Embed 5 ProCohere · $0.09/M tokensNo public result yet –

Gemini Embedding 2Google · $0.15/M tokensNo public result yet –

mistral-embed-2312Mistral AI · $0.08/M tokensNo public result yet –

Rerank 4 ProCohereNo public result yet –

text-embedding-3-largeOpenAI · $0.1/M tokensNo public result yet –

voyage-4-largeVoyage AI (MongoDB) · $0.09/M tokensNo public result yet

Local

harrier-oss-v1-27bMicrosoft · ~54 GB VRAM74.3% on MTEB(Multilingual, v2)$S

KaLM-Embedding-Gemma3-12B-2511Tencent · ~24 GB VRAM72.3% on MTEB(Multilingual, v2)$S

Qwen3-Embedding-8BAlibaba (Qwen) · ~16 GB VRAM70.6% on MTEB(Multilingual, v2)$A

harrier-oss-v1-0.6bMicrosoft · ~2 GB VRAM69% on MTEB(Multilingual, v2)$A

F2LLM-v2-14BAnt Group (codefuse-ai) · ~28 GB VRAM68.7% on MTEB(Multilingual, v2)A

jina-embeddings-v5 (omni-small)Jina AI (Elastic) · ~4 GB VRAM67% on MTEB(Multilingual, v2)B

nomic-embed-text-v2-moeNomic AI · ~2 GB VRAM57.6% on MTEB(Multilingual, v2)B

snowflake-arctic-embed-l-v2.0Snowflake · ~3 GB VRAM57% on MTEB(Multilingual, v2)B

zembed-1ZeroEntropy · ~8 GB VRAM56.6% on MTEB(Multilingual, v2)C

granite-embedding-311m-multilingual-r2IBM · ~1 GB VRAM56% on MTEB(Multilingual, v2)$–

bge-m3BAAI · ~3 GB VRAMNo public result yet –

bge-reranker-v2-m3BAAI · ~2 GB VRAMNo public result yet –

EmbeddingGemma 2Google · ~2 GB VRAMNo public result yet –

jina-reranker-v3.5Jina AI (Elastic) · ~2 GB VRAMNo public result yet –

Nemotron-3.5-Embed-8BNVIDIA · ~17 GB VRAMNo public result yet –

Qwen3-Reranker-8BAlibaba (Qwen) · ~17 GB VRAMNo public result yet –

zerank-2ZeroEntropy · ~9 GB VRAMNo public result yet

Vector databases & engines

Vector databases, search engines, embedding APIs and serving engines that store and query vectors.

Cloud

OpenRouterOpenRouter · Free tiervia Qwen3-Embedding-8B A

Jina Search Foundation APIsJina AI (Elastic) · Free tiervia jina-embeddings-v5 (omni-small)–

PineconePinecone · Free tier · $20/moNo public result yet –

turbopufferturbopuffer · $16/moNo public result yet

Local

OllamaOllama · Freevia Qwen3-Embedding-8B$S

vLLMvLLM project · Freevia Qwen3-Embedding-8B$S

Text Embeddings InferenceHugging Face · Freevia Qwen3-Embedding-8B$–

llama.cppggml-org · FreeNo public result yet –

ElasticsearchElastic · FreeNo public result yet –

Milvus (Zilliz Cloud)Zilliz · FreeNo public result yet –

QdrantQdrant · FreeNo public result yet –

ChromaChroma · FreeNo public result yet –

pgvectorpgvector (open source) · FreeNo public result yet –

Sentence TransformersHugging Face · FreeNo public result yet –

WeaviateWeaviate · FreeNo public result yet –

LanceDBLanceDB · FreeNo public result yet – FastEmbedQdrant · FreeNo public result yet –

InfinityMichael Feil (open source) · FreeNo public result yet –

LM StudioElement Labs · FreeNo public result yet

Retrieval frameworks

Frameworks and platforms you assemble a retrieval or RAG pipeline in.

Cloud

Nothing tracked yet

Local

–

RAGFlowInfiniFlow · FreeNo public result yet –

LlamaIndexLlamaIndex · FreeNo public result yet –

Haystackdeepset · FreeNo public result yet

Everything, in order

All embeddings & retrieval options in quality order
Tier	Name	Approach	Runs	Rated by	Price	Skill
S	harrier-oss-v1-27b Microsoft	Models	Local	74.3% on MTEB(Multilingual, v2)	~54 GB VRAM	—
S	KaLM-Embedding-Gemma3-12B-2511 Tencent	Models	Local	72.3% on MTEB(Multilingual, v2)	~24 GB VRAM	—
S	Qwen3-Embedding-8B Alibaba (Qwen)	Models	Local	70.6% on MTEB(Multilingual, v2)	~16 GB VRAM	—
S	Ollama Ollama	Vector databases & engines	Local	via Qwen3-Embedding-8B	Free	intermediate
S	vLLM vLLM project	Vector databases & engines	Local	via Qwen3-Embedding-8B	Free	advanced
S	Text Embeddings Inference Hugging Face	Vector databases & engines	Local	via Qwen3-Embedding-8B	Free	advanced
S	OpenRouter OpenRouter	Vector databases & engines	Cloud	via Qwen3-Embedding-8B	Free tier	intermediate
A	harrier-oss-v1-0.6b Microsoft	Models	Local	69% on MTEB(Multilingual, v2)	~2 GB VRAM	—
A	F2LLM-v2-14B Ant Group (codefuse-ai)	Models	Local	68.7% on MTEB(Multilingual, v2)	~28 GB VRAM	—
A	jina-embeddings-v5 (omni-small) Jina AI (Elastic)	Models	Local	67% on MTEB(Multilingual, v2)	~4 GB VRAM	—
A	Jina Search Foundation APIs Jina AI (Elastic)	Vector databases & engines	Cloud	via jina-embeddings-v5 (omni-small)	Free tier	intermediate
B	nomic-embed-text-v2-moe Nomic AI	Models	Local	57.6% on MTEB(Multilingual, v2)	~2 GB VRAM	—
B	snowflake-arctic-embed-l-v2.0 Snowflake	Models	Local	57% on MTEB(Multilingual, v2)	~3 GB VRAM	—
B	zembed-1 ZeroEntropy	Models	Local	56.6% on MTEB(Multilingual, v2)	~8 GB VRAM	—
C	granite-embedding-311m-multilingual-r2 IBM	Models	Local	56% on MTEB(Multilingual, v2)	~1 GB VRAM	—
–	llama.cpp ggml-org	Vector databases & engines	Local	No public result yet	Free	advanced
–	RAGFlow InfiniFlow	Retrieval frameworks	Local	No public result yet	Free	intermediate
–	Elasticsearch Elastic	Vector databases & engines	Local	No public result yet	Free	advanced
–	LlamaIndex LlamaIndex	Retrieval frameworks	Local	No public result yet	Free	advanced
–	Milvus (Zilliz Cloud) Zilliz	Vector databases & engines	Local	No public result yet	Free	advanced
–	Qdrant Qdrant	Vector databases & engines	Local	No public result yet	Free	intermediate
–	Chroma Chroma	Vector databases & engines	Local	No public result yet	Free	intermediate
–	Haystack deepset	Retrieval frameworks	Local	No public result yet	Free	advanced
–	pgvector pgvector (open source)	Vector databases & engines	Local	No public result yet	Free	advanced
–	Sentence Transformers Hugging Face	Vector databases & engines	Local	No public result yet	Free	advanced
–	Weaviate Weaviate	Vector databases & engines	Local	No public result yet	Free	intermediate
–	LanceDB LanceDB	Vector databases & engines	Local	No public result yet	Free	intermediate
–	FastEmbed Qdrant	Vector databases & engines	Local	No public result yet	Free	intermediate
–	Infinity Michael Feil (open source)	Vector databases & engines	Local	No public result yet	Free	advanced
–	bge-m3 BAAI	Models	Local	No public result yet	~3 GB VRAM	—
–	bge-reranker-v2-m3 BAAI	Models	Local	No public result yet	~2 GB VRAM	—
–	Embed 5 Pro Cohere	Models	Cloud	No public result yet	$0.09/M tokens	—
–	EmbeddingGemma 2 Google	Models	Local	No public result yet	~2 GB VRAM	—
–	Gemini Embedding 2 Google	Models	Cloud	No public result yet	$0.15/M tokens	—
–	jina-reranker-v3.5 Jina AI (Elastic)	Models	Local	No public result yet	~2 GB VRAM	—
–	LM Studio Element Labs	Vector databases & engines	Local	No public result yet	Free	beginner
–	mistral-embed-2312 Mistral AI	Models	Cloud	No public result yet	$0.08/M tokens	—
–	Nemotron-3.5-Embed-8B NVIDIA	Models	Local	No public result yet	~17 GB VRAM	—
–	Pinecone Pinecone	Vector databases & engines	Cloud	No public result yet	Free tier · $20/mo	intermediate
–	Qwen3-Reranker-8B Alibaba (Qwen)	Models	Local	No public result yet	~17 GB VRAM	—
–	Rerank 4 Pro Cohere	Models	Cloud	No public result yet	—	—
–	text-embedding-3-large OpenAI	Models	Cloud	No public result yet	$0.1/M tokens	—
–	turbopuffer turbopuffer	Vector databases & engines	Cloud	No public result yet	$16/mo	intermediate
–	voyage-4-large Voyage AI (MongoDB)	Models	Cloud	No public result yet	$0.09/M tokens	—
–	zerank-2 ZeroEntropy	Models	Local	No public result yet	~9 GB VRAM	—

Benchmarks used

MTEB(Multilingual, v2) by MTEB (embeddings-benchmark community, hosted on Hugging Face) (automated tests, Mean (Task): average main score over 131 multilingual tasks (bitext mining, classification, clustering, retrieval, reranking, STS, instruction reranking)) · snapshot 19 Oct 2026
470 models listed and ranked by Borda count; only 87 have results on all 131 tasks and thus a Mean (Task). Most current closed APIs (voyage-4, Cohere Embed 4/5, gemini-embedding-2, text-embedding-3) are incomplete. The board shows each model's zero-shot share (tasks not in its declared training data); -1 means undeclared. Read from the public API behind mteb-leaderboard.hf.space.
MTEB(eng, v2) by MTEB (embeddings-benchmark community, hosted on Hugging Face) (automated tests, Mean (Task): average main score over 41 English tasks (classification, clustering, pair classification, reranking, retrieval, STS, summarization)) · snapshot 19 Oct 2026
344 models listed and ranked by Borda count; 187 have results on all 41 tasks. Several top entries have low declared zero-shot shares (43-58%), i.e. trained on in-domain data. Read from the public API behind mteb-leaderboard.hf.space.

Embeddings & retrieval

Quality vs. setup

Everything, in order

Benchmarks used