Voyage AI - Embedding Models

Basic Information

• Product Name: Voyage AI
• Developer Company: Voyage AI
• Country/Region: USA
• Official Website: https://www.voyageai.com/
• Type: Embedding Model API Provider
• Latest Version: Voyage 4 Series (Released January 2026)
• Previous Generation: Voyage 3 Series

Product Description

Voyage AI specializes in providing state-of-the-art embedding models, consistently surpassing competitors like OpenAI and Cohere in retrieval accuracy. The Voyage 4 Series introduces the industry-first shared embedding space technology and Mixture of Experts (MoE) architecture, significantly reducing service costs while maintaining top-tier performance. Voyage also offers domain-specific models (code, finance, law) and multimodal embedding capabilities.

Core Features/Characteristics

• Voyage 4 Series (January 2026):
• voyage-4-large: MoE architecture, 40% lower service cost compared to equivalent dense models
• voyage-4: General-purpose balanced model
• voyage-4-lite: Lightweight and efficient model
• voyage-4-nano: Open-weight ultra-lightweight model
• Shared Embedding Space: All Voyage 4 models generate compatible embeddings, allowing mixed usage of different models for queries and document embeddings
• Voyage 3 Series:
• voyage-3-large: Surpasses OpenAI v3-large by 9.74% and Cohere v3 by 20.71% on BEIR and MIRACL benchmarks
• voyage-3.5 / voyage-3.5-lite
• Domain-Specific Models: voyage-code-3 (code), voyage-finance-2 (finance), voyage-law-2 (law)
• Multimodal Embeddings: Supports text and rich-content images (charts, PPTs, document screenshots, etc.)
• Contextual Embeddings: Document fragments understand their position within the full document, leading retrieval accuracy by 7-23% over competitors
• Matryoshka Learning and Quantization-Aware Training: Supports smaller dimensions and int8/binary quantization

Business Model

• Pay-as-you-go API: Based on token count
• New Pricing Model (May 2025): Updated pricing structure
• Enterprise Customization: Contact sales for tailored solutions

Target Users

• Developers of RAG applications requiring high retrieval accuracy
• Code search and code understanding scenarios
• Professional retrieval in finance and law industries
• Technical teams needing flexible embedding solutions

Competitive Advantages

• Consistently leading retrieval accuracy (BEIR, MIRACL benchmarks)
• Industry-first shared embedding space
• MoE architecture reduces service costs by 40%
• Contextual embeddings lead competitors by 7-23%
• Domain-specific models (code, finance, law)
• Multi-granularity embedding support (Matryoshka + quantization)

Relationship with OpenClaw Ecosystem

Voyage AI's high-precision embedding models can significantly enhance the retrieval quality of OpenClaw RAG systems. Its domain-specific models (code, finance, law) make OpenClaw agents more accurate in specific domain scenarios. The shared embedding space feature allows OpenClaw to flexibly use different models based on scenarios, achieving the best balance between accuracy, latency, and cost.