Mid-Conversation System Messages (platform.claude.com)

Anthropic has introduced a new feature called mid-conversation system messages in its Claude AI models, allowing developers to dynamically add instructions during a conversation without invalidating the existing prompt cache. Previously, modifying the top-level system field would disrupt the cached prompt, resulting in a costly cache-miss for subsequent interactions. With mid-conversation messages, developers can append system instructions directly at the point they become relevant, ensuring that earlier conversation turns remain cached and unchanged, thereby enhancing efficiency. This development is significant for the AI/ML community as it streamlines interactions, enabling more contextually relevant guidance in real-time and preserving performance during extensive dialogues. The feature supports applications requiring strict compliance with instructions, as system messages take precedence over user inputs when conflicting. Mid-conversation messages are available on multiple Claude models, such as Claude Fable 5 and Claude Opus 5, and are expected to improve user experience by providing operators with greater control over conversation flow without compromising cache efficiency.