This newsletter highlights three major developments in the LanceDB ecosystem: the stable-worldmodel platform using Lance for efficient world model training, a new DuckDB-Lance integration enabling multimodal SQL queries by storing images, embeddings, and metadata in unified tables, and benchmarks showing LanceDB’s S3-backed approach costs approximately $779/month for 100M vectors. The DuckDB extension allows teams to query multimodal data without moving it out of the lakehouse. Benchmarks demonstrate significant savings compared to traditional in-memory vector database architectures like OpenSearch.