Claude Opus 4.8: "a modest but tangible improvement"

This article discusses the release of Anthropic's Claude Opus 4.8, highlighting the company's honest acknowledgment of it being a modest incremental improvement over its predecessor. Key features include enhanced honesty in AI responses, with models trained to avoid unsupported claims and flag uncertainties. Opus 4.8 is four times less likely to allow code flaws to pass unremarked and shows the lowest incorrect-rate on benchmarks by abstaining on uncertain questions. The article also covers pricing details, a reduced-cost fast mode for research preview organizations, and a notable new feature allowing mid-conversation system messages for more efficient long-running interactions.