Cosmos DB for NoSQL vector index: flat vs quantizedFlat vs diskANN

This sheet compares flat, quantizedFlat, diskANN.

CriterionflatquantizedFlatdiskANN
Best whenTiny sets needing exact recall, up to 505 dimensionsFiltered search over a small vector set needing high accuracySearch scoped past ~50,000 vectors, lowest latency and RU

The verdict, the full comparison, 5 rules and 3 traps are part of AI-200 access. Unlock AI-200.