Anthropic has introduced "Dreaming" for its Claude Managed Agents, a new feature that allows agents to review past sessions and memory stores to identify patterns and refine their long-term memory. This capability is currently in a research preview and is exclusive to Anthropic-hosted Managed Agents, not the bare Messages API. The company also announced that "outcomes" and "multi-agent orchestration" are moving from research preview to broader availability. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Enhances agent memory management, potentially improving long-term task performance and reducing repetitive errors in asynchronous, long-running agent operations.
RANK_REASON This is a product feature announcement for an existing model, not a new model release.