Meta's Context Language Models treat context as an editable file, +65% on agent tasks

arankomatsuzaki · x · 2026-09-30

Meta (with UW, MIT, Trillium Labs) introduces Context Language Models (CLMs), which natively manage their own context by treating it as an editable file the model can freely update, instead of an append-only conversation history. This lets the model learn what to keep in context and naturally extends to multi-agent systems where each agent's context is a file. Zero-shot CLMs beat SOTA context-management strategies: +65% score at equal compute on a 24-hour multi-repo agent-swarm task, +11.4% accuracy with 21.5% fewer FLOPs on BrowseComp-Plus.

Related event: Meta Introduces Context Language Models (CLM), Open-Sourced(3 posts)→

Original post →

More from coding & agent

coding & agent channel →