Introduction

Benchmark Datasets

Cohere Wikipedia Embeddings (100K)

Million Text Embeddings

ArXiv Embeddings (Ada-002)

DBpedia Entity Embeddings (OpenAI 1M)

Benchmark Configuration

HNSW Index Parameters

Qdrant-Specific Configuration

Cosdata Configuration

Raw Performance Results

Qdrant Performance Results (With Scalar Quantization)

Cosdata Performance Results

Head-to-Head Comparison: Cosdata vs Qdrant (Both Quantized)

Performance Results Analysis

Query Performance Comparison

Qdrant Performance Profile

Cosdata Performance Profile

Throughput Analysis

Queries Per Second (QPS)

Data Insertion Performance

Recall Accuracy Trade-offs

Key Performance Insights

Cosdata's Competitive Advantages

Superior Data Ingestion Performance

Higher Throughput With Default Quantization

Superior Query Performance

Optimized Default Configuration

Consistent Performance Scaling

Qdrant's Strengths

Consistent Performance Profile

Perfect Recall Accuracy

Architectural Implications

Cosdata's Comprehensive Approach

Qdrant's Specialized Optimization

Conclusion

Technical Specifications

Test Environment

Reproducibility