Skip to content

Anthropic studies how Claude remembers across sessions

What happened

AI digest

Anthropic has published research explaining how Claude retains knowledge across sessions without changing the model, with the central finding that selective note-taking beats archiving everything. In a Pokemon Red test, Claude Sonnet 3.7 created 413 files and 225 folders in about 10,000 steps but ended up with only one badge; Claude Fable 5 kept just a few files, recording strategy and verified rules, and won four badges. The research argues that the key to agent memory is knowing what to forget.

Written by AI from the reports · updated 10 h ago

Timeline

Follow the reports to see the event from each side.

Nov 16
  1. The Decoder
    Anthropic says the key to agent memory is knowing what to forget, not how much to store

    Anthropic studied how Claude retains knowledge across sessions. Selective notes beat exhaustive archives by a wide margin. In a Pokémon Red test, the newer Claude Fable 5 got four times as far as the older Sonnet 3.7 by keeping just a handful of strategy files instead of more than 400. A "Dreaming" process retroactively corrects faulty memories.

Heat of this event

Current heat 7·Peak in the comparable range 10(Nov 17 01:00)·Change over 24 hours in the comparable range –

02.557.510Nov 1701:00Nov 1704:00Nov 1707:00Nov 1710:00

The trend compares only the same accounts observed without a break, so its range can be narrower than the current heat count. Move the pointer or tap the chart to see each hour; with a keyboard, use the left and right arrow keys.