Anthropic studies Claude's cross-session memory mechanism
Understand this story
Anthropic published a research article stating that the key to Claude's cross-session memory lies in selective recording rather than archiving the entire process; the model itself remains unchanged and relies on notes to retain knowledge between sessions. In the Pokémon Red test, Claude Sonnet 3.7 generated 413 files and 225 folders over about 10,000 steps, ultimately earning only 1 badge; Claude Fable 5 recorded only strategies and verified rules, earning 4 badges. Anthropic said the accompanying Dreaming error-correction mechanism was launched as a research preview in May for its AI agent cloud platform Claude Managed Agents, and at launch supported only Claude Opus 4.7 and Sonnet 4.6.
AI overview based on source reports · Updated 1 days ago
Coverage timeline
Follow the coverage to explore different perspectives.
- The DecoderAnthropic says the key to agent memory is knowing what to forget, not how much to store
Anthropic published research on how Claude retains knowledge across sessions, finding that selective notes outperform exhaustive archives and that a "Dreaming" process, in which Claude reviews transcripts of earlier sessions to check its notes, can correct errors. In a Pokémon Red test, Claude Sonnet 3.7 created 413 files across 225 folders and earned one badge, while Claude Fable 5 kept to a few files and earned four badges. Anthropic says open questions remain, including that note-taking distracts the agent, Dreaming costs extra compute, and memories need regular upkeep.
Activity for this story
- Current comparable activity
- 4
- Peak comparable activity
- 10Nov 17 01:00
- Change in the past 24 hours
- -50%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.