Mini toolsVoyage models
Browse all models from this model family.
ID
Model
Company
Type
Primary task
Open source
-
By Voyage AIvoyage-4-large is a state-of-the-art general-purpose and multilingual text embedding model using a mixture-of-experts (MoE) architecture. It delivers frontier retrieval accuracy with serving costs 40% lower than comparable dense models. Features a 32K context window, flexible output dimensions (256, 512, 1024, 2048), and multiple quantization options including float, int8, uint8, binary, and ubinary.TextReleased 5mo ago
-
By Voyage AIMultimodal embedding model that vectorizes interleaved text, images, and videos into a shared vector space for retrieval tasks. Uses a single transformer encoder for all modalities, eliminating the modality gap. Supports document screenshots, PDFs, slides, figures, tables, and video frames. Context window of 32,000 tokens with Matryoshka embeddings at 256, 512, 1024, and 2048 dimensions.MultimodalReleased 5mo ago
-
By Voyage AILightweight, general-purpose text embedding model optimized for low latency and cost. Supports a 32K token context window and flexible output dimensions (2048, 1024, 512, 256) via Matryoshka learning. Offers multiple quantization formats including float, int8, uint8, binary, and ubinary. Part of the Voyage 4 series with a shared embedding space across models.TextReleased 5mo ago
-
By Voyage AIvoyage-4-nano is a lightweight open-weight text embedding model from the Voyage 4 series. It features a 32,000-token context window, 340M total parameters, multilingual support, Matryoshka flexible dimensions (256 to 2048), quantization-aware training, and a shared embedding space compatible with all other Voyage 4 models. Licensed under Apache 2.0.TextReleased 5mo ago
-
By Voyage AIText embedding model optimized for general-purpose and multilingual retrieval quality. Supports a 32,000-token context window and configurable output dimensions (256, 512, 1024, 2048) via Matryoshka Representation Learning. Part of the Voyage 4 series, which shares a common embedding space enabling asymmetric retrieval across models in the family.TextReleased 5mo ago
-
By Voyage AIvoyage-context-3 is a contextualized chunk embedding model that produces vectors capturing both chunk-level detail and full document context. It processes entire documents in a single pass and generates a distinct embedding per chunk. Supports flexible output dimensions (256, 512, 1024, 2048) via Matryoshka learning and multiple quantization options (float32, int8, uint8, binary). Per-chunk context window is 32K tokens; total document context is 120K tokens.TextReleased 11mo ago
-
By Voyage AIText embedding model optimized for general-purpose and multilingual retrieval across law, finance, code, and long-document domains. Supports a 32K-token context window, flexible output dimensions (256, 512, 1024, 2048), and multiple quantization formats (float32, int8, uint8, binary) via Matryoshka and quantization-aware training. Accessed via API.TextReleased 1y ago
-
By Voyage AIText embedding model optimized for code retrieval. Supports Matryoshka embeddings at 256, 512, 1024, and 2048 dimensions with quantized formats (float, int8, uint8, binary, ubinary) for up to 32x storage reduction. Features a 32K-token context window. Outperforms OpenAI text-embedding-3-large by 13.80% on 32 code retrieval datasets covering text-to-code, code-to-code, and docstring-to-code tasks.CodingReleased 1y ago
-
By Voyage AIMultimodal embedding model that vectorizes interleaved text and images through a unified transformer encoder. Supports screenshots of PDFs, slides, tables, and figures without complex document parsing. Unlike CLIP-based models, eliminates the modality gap, enabling accurate mixed-modality retrieval across text and visual content.MultimodalReleased 1y ago
-
By Voyage AILightweight general-purpose text embedding model optimized for latency and cost. Generates 512-dimensional vector embeddings with a 32K-token context window. Outperforms OpenAI v3 large by 3.82% on average retrieval accuracy across eight domains while costing 6.5x less, at $0.02 per 1M tokens.TextReleased 1y ago
-
By Voyage AIText embedding model optimized for multilingual retrieval and RAG. Supports 27 languages including French, German, Japanese, Spanish, Korean, and English, with a 32K token context window and 1024-dimensional embeddings. Outperforms OpenAI text-embedding-3-large and Cohere multilingual v3 on average across evaluated languages.TextReleased 2y ago
-
By Voyage AIFinance domain-specific text embedding model optimized for financial retrieval and RAG applications. Supports a 32K context window and outputs 1024-dimensional vectors. Outperforms general-purpose embedding models across 11 financial datasets including SEC filings, earnings reports, financial news, and hybrid tabular/text question-answering benchmarks.TextReleased 2y ago
-
By Voyage AIvoyage-large-2-instruct is a general-purpose text embedding model with a 16K context window and 1024-dimensional output vectors. It is instruction-tuned for enhanced performance on retrieval, classification, clustering, and reranking tasks. At release it ranked #1 on the MTEB leaderboard with an average score of 68.28, outperforming OpenAI v3 large and Cohere English v3.TextReleased 2y ago
-
By Voyage AIText embedding model optimized for legal retrieval and RAG. Trained on 1 trillion legal tokens with a 16,000-token context window, it tops the MTEB legal retrieval leaderboard, outperforming OpenAI text-embedding-3-large by 6% on average across 8 legal datasets and by over 10% on three of them. Outputs 1024-dimensional embeddings.TextReleased 2y ago
-
By Voyage AIvoyage-code-2 is an embedding model optimized for semantic code retrieval. It supports a 16,000-token context window and produces 1536-dimensional vector embeddings. Trained on large-scale code datasets using advanced loss functions and contrastive learning, it outperforms competing models on code retrieval benchmarks and also improves on general-purpose text retrieval tasks.CodingReleased 2y ago
-
By Voyage AIvoyage-lite-02-instruct is a lightweight instruction-tuned text embedding model from Voyage AI's second generation. Ranked #3 on MTEB at launch, it achieves top-tier accuracy with 6x fewer parameters and 4x smaller embedding dimensions than other top-5 models. Supports a 4,096-token context window and task-specific instruction prefixes to boost performance across retrieval, classification, and semantic search.TextReleased 2y ago
-
By Voyage AILightweight text embedding model fine-tuned on top of voyage-lite-01 for classification and clustering tasks. Features a 4096-token context window and 1024-dimensional output vectors. Not recommended for retrieval or search; classification and clustering are the only recommended use cases.TextReleased 2y ago
-
By Voyage AIvoyage is a text embedding model that converts text into dense vector representations for semantic retrieval. Originally released as voyage-01 via the Voyage AI API, this Hugging Face repository contains only the tokenizer (MIT licensed). Built on a LlamaTokenizer architecture with a 4,096-token context window, designed to outperform OpenAI text embeddings on retrieval benchmarks.TextReleased 2y ago
-
By Voyage AIvoyage-01 is a text embedding model by Voyage AI, designed for retrieval-augmented generation (RAG) and semantic search. It maps text inputs to dense vector representations for document and query retrieval tasks. Deprecated in favor of newer Voyage models.TextReleased 2y ago
-
By Voyage AILightweight text embedding model from Voyage AI for retrieval-augmented generation and semantic search. Converts text inputs into dense vector representations with a 4096-token context window. A smaller, faster variant of voyage-01, now deprecated and superseded by newer Voyage AI embedding generations. Available via API and as open weights under MIT license.TextReleased 2y ago
