Research2h ago
RENDER Benchmark Tests LLM Memory Formatting
Matched-budget resolved packets outperformed recency-truncated raw dialogue by 42.4 to 72.6 points in new tests. The RENDER benchmark varies how memory is presented—as summaries, typed records, or raw excerpts—while keeping conversation history constant. Results across nine models show that formatting drastically alters retrieval accuracy. Developers should optimize memory rendering over simple truncation.