Anthropic launches Opus 4.8, with honesty as its killer feature (www.zdnet.com)

Anthropic has launched Claude Opus 4.8, introducing a significant focus on "honesty" as a key feature of this latest large language model. This update enhances Opus's reliability by reducing the chances of it making unsupported claims and increasing its ability to express uncertainty when answers are not clear. In practical terms, evaluations show Opus 4.8 is four times less likely to overlook flaws in code it generates compared to its predecessor. Users have reported that it demonstrates more effective judgment, asking pertinent questions and catching its own mistakes, which significantly improves the user experience in coding and complex task management. One of the standout features of Opus 4.8 is its dynamic workflows capability, allowing the execution of hundreds of parallel subagents that can adapt tasks in real time based on evolving project needs. This development is particularly valuable for large-scale coding tasks, such as extensive codebase migrations. The model is designed to verify outcomes before reporting back to users, ensuring that the outputs are reliable, especially as it coordinates numerous subagents. Available through the Claude platform and API, Opus 4.8 retains the pricing structure of its predecessor while introducing a faster, more cost-effective mode for users who need quicker results.