Voyage 3

First fully reproducible open-source text embedding model with 8,192 context length. v2 introduces Mixture-of-Experts architecture for multilingual embeddings. Outperforms OpenAI models on benchmarks. This is an OSS model under Apache 2.0 license.

Multimodal embedding model from Alibaba's Qwen family that processes text, images, and visual documents in a unified embedding space for cross-modal retrieval tasks.

State-of-the-art ColBERT-style embedding model family achieving top performance on ViDoRe benchmarks for visual document retrieval. The 8B model ranks first on ViDoRe V3 leaderboard with 63.42 average NDCG@10 as of February 2026.

Meta's groundbreaking multimodal embedding model that learns a joint embedding space across six modalities (images, text, audio, depth, thermal, IMU) using only image-paired data, enabling cross-modal retrieval and zero-shot capabilities.

Frontier multilingual text embedding model with 570M parameters and 8192 token-length, featuring task-specific LoRA adapters and outperforming OpenAI and Cohere embeddings on MTEB benchmark.

Microsoft's state-of-the-art multilingual text embedding model supporting 100 languages with 1024-dimensional embeddings, trained on 1 billion multilingual text pairs for robust cross-lingual retrieval.