Anthropic releases Opus 4.8 with focus on AI honesty

Anthropic has released Opus 4.8, the newest version of its most advanced publicly available AI model, available at the same price as its predecessor. The release comes just 41 days after Opus 4.7, a notably faster upgrade cycle driven partly by user disappointment with the previous version and increased competitive pressure from OpenAI's Codex and Google's Gemini Flash releases. The updated model places particular emphasis on honesty and uncertainty management, addressing a common AI weakness where systems confidently make unsupported claims. Early testers at Bridgewater Associates noted that Opus 4.8 proactively flags issues with inputs and outputs that other models typically miss. Alongside the new model, Anthropic launched Dynamic Workflows, a research preview feature designed to help advanced models manage complex tasks across hundreds of parallel subagents, enabling Opus to handle codebase-scale migrations from start to finish. Anthropic's most powerful model, Mythos, which features advanced cybersecurity capabilities that have raised concerns among executives and world leaders, remains under wraps pending additional safeguards. The company indicated that under Project Glasswing, partners Amazon, Microsoft and Apple will have access to Mythos for cybersecurity purposes. Anthropic expects to bring Mythos-class models to all customers within the coming weeks.