Qdrant 1.19
#1 en Embeddings & bases vectoriellesQdrant · v1.19 · 2× · vu le 13 août 2026
100
Momentum
Qdrant 1.19 is an Open Source release of the vector database Qdrant, published on August 4, 2026. The version introduces the new storage format "Turbo4", which stores vectors only as 4-bit TurboQuant compression without full-text copies, thereby saving storage space and increasing throughput. Additional features include a unified storage tier parameter (pinned/cached/cold), prefix matching on keyword indices, and a new slice filter condition for deterministic subsets. Qdrant remains Apache-2.0-licensed, self-hostable, and available as a managed cloud service, with identical engine and API across all deployment modes.
Historique du momentum
15.05.13.08.
Fonctionnalités
| Throughput/Latency | Turbo4 datatype (4-bit compression, no full-precision copy) reduces reads/writes and improves throughput, especially for multi-vector/ColBERT search |
| License | Open source, Apache 2.0 |
| Platform | Docker, Kubernetes (Helm/Operator), binary; clients for Python, JS/TS, Rust, Go, Java, .NET |
| Price | Self-hosted: free (Apache 2.0); Qdrant Cloud: free tier + usage-based pricing from ~$0.014/hr, typ. $100–$500/month |
| Protocol Compatibility | REST API (OpenAPI 3.0) and gRPC interface, each REST endpoint has a corresponding gRPC method |
| Release Date | August 4, 2026 (v1.19.0) |
| Supported Models/Providers | Supports dense, sparse (BM25/IDF) and multi-vectors (e.g. ColBERT late-interaction); cloud inference for selected embedding models |