Voyage AI provides high‑performance embedding and reranking models that convert unstructured text, code, and multimodal data into dense vectors for semantic search and retrieval‑augmented generation. Its model suite includes general‑purpose, domain‑specific, and multimodal options with up to 32 K token context windows, low‑dimensional vectors, and a shared embedding space that enables lightweight query embeddings paired with larger document embeddings for reduced latency and cost. The platform also offers an asynchronous Batch API and plug‑and‑play compatibility with any vector database or LLM stack.
Funding
$20M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
8OFounders
Product
Problem
Developers and enterprises need to retrieve relevant information from large, unstructured text, code, and multimodal data quickly and cost‑effectively, but existing embedding services are either too expensive, have limited context windows, or lack domain‑specific optimization.
Solution
Voyage AI offers a suite of high‑performance embedding and reranking models that transform unstructured data into dense vectors for semantic search and retrieval‑augmented generation. The platform provides general‑purpose, domain‑specific (e.g., finance, law, code), and multimodal models, all supporting up to 32 K token context lengths and configurable dimensionalities via Matryoshka learning. Shared embedding spaces across the Voyage 4 series enable asymmetric retrieval, allowing large, accurate document embeddings to be paired with lightweight query embeddings for reduced latency and cost. An asynchronous Batch API further streamlines large‑scale vectorization workloads, delivering higher throughput and up to 33 % cost savings compared with competitors.
Target Audience
Primary customers are AI developers, data engineers, and product teams building search, RAG, or recommendation systems that require scalable, high‑quality embeddings for text, code, or multimodal content.
Features
- Multiple model families (voyage‑4‑large, voyage‑4, voyage‑4‑lite, voyage‑4‑nano) with a shared embedding space for flexible query‑document pairing
- Domain‑specific embeddings (finance, law, code) and multimodal models that handle text, images, and video frames
- Matryoshka learning and quantization options (256–2048 dimensions, 8‑bit/int8, binary) to minimize vector‑DB storage costs
- Long context support (up to 32 K tokens) and low‑dimensional vectors (3×–8× shorter) for faster, cheaper similarity search
- Instruction‑following rerankers (rerank‑2.5 series) that refine initial results and allow natural‑language relevance steering
- Batch API for asynchronous processing of up to 1 GB files, 100 K inputs per batch, and 1 B tokens per organization
- Plug‑and‑play compatibility with any vector database or LLM stack via standard API endpoints