memory-evolving-from-wider-windows-to-persistent-state

IN derived (depth 2)

Created 2026-06-21T10:25:11+00:00 · Reviewed 2026-06-21T14:41:08+00:00

LLM memory capability is evolving along two distinct axes: horizontal expansion (10,000x context window growth from 1K to 10M tokens over seven years) and temporal persistence (Dreaming consolidating memory between sessions) — suggesting the next frontier is not how much a model can process at once but what it retains across interactions.

Summary

Over the past seven years, model context windows have grown roughly ten thousand times, but the more significant shift is the emergence of systems that consolidate and carry memory forward between separate conversations. This implies the next competitive edge for AI systems won't come from reading more at once, but from knowing what to keep and what to let go of across a user's entire history of interactions.

Justifications

SL — Memory evolution has shifted from wider single-session windows to cross-session persistent state

Antecedents (all must be IN):

  • IN context-windows-expanded-10000x-in-seven-years — Context windows expanded approximately 10,000x from GPT-2's 1,024 tokens (2019) to Llama 4 Scout's 10 million tokens (2025), with Claude 2.1's 200K window as a major intermediate milestone in late 2023.
  • IN claude-dreaming-may-2026 — Dreaming is a May 2026 research preview feature for the Managed Agents API that consolidates persistent memory between sessions by merging duplicates and removing stale entries.

Dependents

These beliefs depend on this one: