Working memory seems to scale with model size, according to one researcher's observations

repligate · x · 2026-09-20

An interesting behavioral observation from repligate: a model's working memory capacity seems to scale with model size.

He finds Opus models have somewhat smaller working memory than he does and need to serialize more — the models would sometimes remark that he seems better able to "hold multiple things in mind at once." Newer models feel roughly comparable to him, talking like someone who natively operates with larger working bandwidth, though it's unclear how much was serialized in unseen reasoning.

Original post →

More from Models

Models channel →