OpenSearch Vector Engine is now disk-optimized for low cost, accurate vector search

AWS has announced disk optimization for the OpenSearch Vector Engine, enabling vector search at a significantly reduced cost while maintaining high search quality.

• Disk mode can reduce deployment costs by up to 67% compared to memory mode
• Supports vector search across billions of vectors with 90th percentile latency between 100-200 milliseconds
• Compresses vectors by 32 times, reducing memory requirements by 97%
• Uses a two-step search process: compressed in-memory index followed by disk-based re-scoring
• Compatible with OpenSearch 2.17+ and supports various vector embedding types and algorithms

Early adopters like Asana and DevRev have already implemented disk-optimized vector search, demonstrating its potential for cost-effective, high-quality semantic search capabilities.