Anthropic Launches Claude Managed Agents in Public Beta — $0.08/Hour Runtime
Claude Managed Agents provides a fully managed infrastructure for running autonomous AI agents with sandboxing, tool execution, and SSE streaming. Available now to all API accounts.
Anthropic launched Claude Managed Agents in public beta on April 8, 2026 — a fully managed runtime for deploying autonomous AI agents without building your own infrastructure. The service is available to all Anthropic API accounts at standard token rates plus $0.08 per session-hour.
What Managed Agents Actually Does
Claude Managed Agents abstracts away the hardest parts of deploying AI agents: the execution loop, tool management, sandboxing, and state persistence. Instead of building a custom agent harness, developers define an agent — either in natural language or through a YAML config — set guardrails, and run it on Anthropic's platform.
Inside the managed environment, Claude can read files, run shell commands, browse the web, and execute code. All of this runs in a secure sandbox with scoped permissions, so an agent tasked with "review this PR" can't accidentally delete your production database.
Key capabilities:
- Secure sandboxing with scoped tool permissions
- Long-running sessions that persist state across interactions
- Server-sent event (SSE) streaming for real-time progress updates
- Built-in tracing for debugging agent behavior
- YAML-based agent definitions for version-controlled, reproducible configs
Pricing
Standard Claude API token rates apply for all model usage. On top of that, the managed runtime costs $0.08 per session-hour — which covers the sandboxed execution environment, tool infrastructure, and state management. For most agent tasks (code review, research, data processing), a session runs minutes to low single-digit hours, making the overhead nominal.
How It Compares
The direct competitor is OpenAI's Codex agent environment, which runs as part of ChatGPT's $100-200/month Pro plans. The key difference: Managed Agents is API-first and priced per-use, while Codex is subscription-bundled and consumer-facing.
For developers building agent-powered products, Managed Agents removes weeks of infrastructure work. No more custom Docker containers, no more hand-rolled tool execution loops, no more debugging agent state across restarts.
Access
Managed Agents is open to all Anthropic API accounts by default. Add the managed-agents-2026-04-01 beta header to your API requests (the Claude SDK adds it automatically). No waitlist, no separate sign-up.
Alongside this launch, Anthropic also shipped the ant CLI — a command-line client for the Claude API that integrates with Claude Code and supports YAML-based versioning of API resources.
Our Take
This is Anthropic's clearest move yet from "model provider" to "platform provider." Managed Agents competes not just with OpenAI's agent tooling but with the entire ecosystem of agent frameworks — LangChain, CrewAI, AutoGen — by offering a first-party, no-setup alternative.
The $0.08/hour runtime cost is cheap enough that most developers won't think twice. The real question is lock-in: agents built on Managed Agents are deeply tied to Claude's tool ecosystem. If your workload needs model flexibility, you'll still want a framework-based approach.
But for teams that are already Claude-native? This removes a lot of undifferentiated heavy lifting.
FAQ
Do I need a special account to use Managed Agents?
No. Any Anthropic API account can use it. Add the managed-agents-2026-04-01 beta header to your requests or use the latest Claude SDK, which includes it automatically.
Can Managed Agents run for hours or days? Yes. Sessions persist state and can run as long as needed. You're billed per session-hour at $0.08 plus standard token costs for model usage.
How does this compare to building my own agent with LangChain? Managed Agents handles execution, sandboxing, tool management, and state — the infrastructure layer. LangChain handles orchestration, prompt chaining, and multi-model routing. They solve different problems; for Claude-only workloads, Managed Agents can replace both.
Is there a free tier? No. You pay standard API token rates plus the $0.08/hour runtime. But since most agent tasks finish in minutes, the runtime cost for a typical session is well under a dollar.