Mem0 Memory Compression Engine: 90% Token Reduction with Graph-Enhanced Knowledge Graphs — 26% Accuracy Gain Over OpenAI on LOCOMO

Mem0·high signal

Mem0's memory compression architecture reduces prompt tokens from ~26K to ~1.8K per conversation (90% reduction) while delivering 26% relative accuracy gains over OpenAI on the LOCOMO benchmark with 91% lower p95 latency. The graph-enhanced variant Mem0g builds a directed labeled knowledge graph alongside the vector store during extraction, with entity extraction, relation generation, and conflict detection preventing contradictory information from corrupting agent memory.