Anyone benchmarked context recall on Claude Opus (Opus 4.5) vs the last version?
Anyone benchmarked context recall on Claude Opus (Opus 4.5) vs the last version?
Doing a large migration and Claude Opus (Opus 4.5) keeps 're-discovering' files it already read this session, re-reading them instead of trusting its own earlier summary. Feels like effective context is smaller than advertised.
Nobody has tried to reproduce this yet. Be the first to confirm it — or report something different.