Research3h ago
RENDER Benchmark Tests LLM Memory Formatting
Matched-budget resolved packets beat recency-truncated raw dialogue by up to 72.6 points in the new RENDER benchmark. Researchers tested nine models across 500 questions to see how input formatting affects memory retrieval. The study proves that LangChain summaries and typed records significantly alter accuracy. Developers must optimize prompt templates to prevent memory loss.