Mnemo Memory — Benchmarks (LongMemEval-S)

Mnemo Memory scores 80.8% strict (404/500) on the full LongMemEval-S benchmark under an OpenAI gpt-4o-mini judge — US-hosted and model-agnostic. The Search API runs a consistent ~150ms warm (p50) on live prod /v1/search. Run date: June 13, 2026.

• single-session-user: 89% (62/70)
• single-session-assistant: 88% (49/56)
• knowledge-update: 85% (66/78)
• temporal-reasoning: 80% (107/133)
• multi-session: 76% (101/133)
• single-session-preference: 63% (19/30) — current bottleneck

Two distinct latencies, not to be conflated: SEARCH (the /v1/search product) is a consistent ~150ms warm p50; ANSWER generation (optional LLM synthesis on top) is 3.6s median in fast mode. Retrieval path: embed the query (text-embedding-3-small) then hybrid search in parallel (pgvector HNSW vector + Postgres full-text keyword + fact-aware), fused with RRF, returning ranked memories with citations.