Anthropic 推出 Claude Managed Agents 公测版

In today’s issue:

• Anthropic launches Claude Managed Agents
• Meta ships Muse Spark multimodal model
• Claude Mythos powers Project Glasswing
• Advisor strategy pairs Opus with Sonnet
• GLM-5.1 tops open-source coding benchmarks
• Microsoft open-sources Memento
• Claude Code ships Monitor tool
• AXI outperforms MCP on browser tasks
• SAGE evolves four-agent reasoning loops
• Self-organizing agents outperform fixed structures

And all the top AI dev news, papers, and tools.

Top Stories

Claude Managed Agents

Anthropic launched Claude Managed Agents in public beta, a suite of composable APIs for building and deploying cloud-hosted agents at scale. The platform pairs a tuned agent harness with production infrastructure, letting developers go from prototype to launch in days instead of months.

• Production-grade sandboxing: Managed Agents handles secure execution, authentication, tool orchestration, and persistent progress for agents that operate autonomously for hours, removing the infrastructure burden from development teams.
• Multi-agent coordination: A research preview enables agents to direct other agents, opening up hierarchical delegation patterns where a planning agent can spin up and manage specialized worker agents.
• Self-evaluation loops: Agents can iterate toward defined success criteria using built-in evaluation capabilities, improving structured file generation task success by up to 10 percentage points on complex problems.
• Enterprise adoption: Notion, Asana, Sentry, Rakuten, and Vibecode are already shipping production agents on the platform, each built in under a week using the managed infrastructure.

Blog

Read more