Models

Most popular open-source multilingual embedding model. Supports dense, sparse, and multi-vector retrieval.

Multilingual multimodal embedding model for 100+ languages. Leading multilingual retrieval benchmarks.

First natively multimodal embedding model supporting text, images, video, audio, and documents. 3072 dimensions, 100+ languages.

Multilingual text embedding model with Matryoshka representation learning.

Multimodal embedding model supporting text, images, and PDFs. Built on Qwen2.5-VL-3B with LoRA adapters.

Compact Qwen3 embedding for edge and low-resource deployment. Apache 2.0.

Mid-size Qwen3 embedding balancing performance and efficiency. Apache 2.0.

#1 on MTEB multilingual leaderboard. Best open-source embedding model. Apache 2.0.

Previous generation Voyage model with strong retrieval performance.

Top-performing embedding model using MoE architecture. Outperforms OpenAI and Cohere on retrieval benchmarks.

Lightweight Voyage 4 variant for cost-sensitive workloads.

OpenAI's most capable embedding model. 3072 dimensions, strong across retrieval, classification, and clustering.

Cost-efficient embedding model balancing performance and price. 1536 dimensions.