Hindsight Benchmarks

Industry benchmarks and model leaderboard for Hindsight — the memory layer for AI agents.

Industry Benchmarks

Benchmark

LongMemEval

Comprehensive long-term memory evaluation across 500 items.

LoComo

Long conversation memory benchmark across real-world dialogues.

Model Leaderboards

Retain Leaderboard

Ranked LLMs for retain() and observation consolidation — fact extraction quality, speed, cost, and reliability.

Reflect Leaderboard

Ranked LLMs for the reflect operation in Hindsight.

Reranker Leaderboard

Ranked rerankers for recall() — which reranker surfaces the most relevant facts first.

Embeddings Leaderboard

Ranked embedding models for Hindsight — affects both retain() storage and recall() retrieval quality.