Anthropic releases Claude Opus 4.8 with improved honesty

anthropic launched claude opus 4.8 with improvements focused on 'honesty' and uncertainty detection. the company states the new model is four times less likely than its predecessor to let flawed code pass without notice and avoids making unsupported claims. users can now control the effort level for responses to manage token usage, trading accuracy for speed. additionally, a 'dynamic workflows' research preview enables the model to plan tasks and run hundreds of parallel subagents in a single session.