Anthropic says Opus 4.8 is "around four times less likely than its predecessor to allow flaws in code it has written to pass unremarked", meaning the model is better at flagging its own uncertainty rather than confidently shipping broken work. The alignment assessment, included in the system card, says the model shows "substantially lower" rates of misaligned behavior than Opus 4.7. Two new features land alongside the model. The first is an effort control slider in Claude.ai and Cowork that lets users choose how hard the model thinks about a given prompt. Higher effort settings spend more tokens for better answers; lower settings respond faster and consume less of a user's rate limit. The second is dynamic workflows, a Claude Code feature that lets the model plan and run hundreds of parallel subagents in a single session, with output verification before reporting back. Anthropic's example use case: codebase-scale migrations across hundreds of thousands of lines of code, from kickoff to merge.