voyage-code-3 Embedding Model
Text embedding model for code search & AI apps. Supports 32K context, multiple output sizes, and quantized embeddings.
ExploreProduct Description
Overview
Text embedding models convert text into numerical vectors and are essential for powering semantic search, retrieval systems, and retrieval-augmented generation (RAG), directly influencing retrieval accuracy. voyage-code-3 is specifically fine-tuned for code retrieval, outperforming OpenAI-v3-large and CodeSage-large by 13.80% and 16.81% across 238 code retrieval datasets. Using Matryoshka learning and support for compact formats like int8 and binary, it reduces storage and search costs while keeping performance high. It delivers 90 ms latency per query (≤100 tokens) and scales to 12.6M tokens/hour at $0.22 per 1M tokens on an ml.g6.xlarge.
More info: https://blog.voyageai.com/2024/12/04/voyage-code-3/
Highlights
Code-optimized, outperforming OpenAI-v3-large and CodeSage-large by 13.80% and 16.81% on 238 retrieval benchmarks.
Supports embedding sizes of 2048, 1024, 512, and 256, with multiple quantization options:
float (32-bit), int8, uint8, binary (compressed int8), and ubinary (compressed uint8).32K token context window, with 90 ms query latency (≤100 tokens) and 12.6M tokens/hour throughput at $0.22 per 1M tokens on ml.g6.xlarge.