Overview
- Anthropic's Claude Managed Agents, launched Wednesday in public beta, charges standard model token rates plus $0.08 per active session-hour on the Claude Platform.
- The service provides a managed agent harness that covers sandboxed code execution, checkpointing, credential and permission controls, and end-to-end tracing.
- Agents run on Anthropic’s own cloud in isolated containers with context persistence and error recovery, and the service is not currently available through Amazon Bedrock or Google Vertex AI.
- Advanced capabilities remain in research preview, including multi-agent orchestration, richer memory, and self-evaluation loops, and Anthropic reports up to a 10-point task success lift in internal tests versus basic prompting.
- Early customers like Notion, Rakuten, Asana and Sentry report production use, and Anthropic is widening its enterprise push with Claude Cowork now generally available and annualized revenue above $30 billion.