CLAUDE OPUS 4.8: “A MODEST BUT TANGIBLE IMPROVEMENT”

Summary from the TLDR newsletter

Claude Opus 4.8's most prominent improvement is its ability to be honest. It is more likely than previous models to flag uncertainties about its work and less likely to make unsupported claims. The model has a new feature that allows mid-conversation system messages. This means users can append updated instructions later in long-running conversations without restating the full system prompt.