Qdrant 1.20 adds native BM25 and TurboQuant quantization

Qdrant v1.20 brings BM25 scoring directly into payload text indexes, new turbo8 and turbo16 vector types that cut storage by 60–75%, and a code-aware tokenizer. The release lets teams drop external sparse models and separate full-text engines from hybrid search pipelines.

• BM25 is computed in-engine during index scans, no external sparse models
• TurboQuant turbo8 and turbo16 shrink vector storage by 60–75%
• New tokenizer preserves semantics of identifiers like parse_ast_node
• Nested payload indexes speed up filtering on deep JSON metadata