Beta requirement
Managed Agents is still beta and requires the managed-agents-2026-04-01 beta header on API requests. The SDK adds it automatically.
Pricing & Limits
Claude Managed Agents pricing and limits
Use this page when you need the operational limits and runtime envelope before you promise capacity or security posture.
Last updated
2026-04-12
What Anthropic Documents
The first limits you need to plan around are beta, rate, and spend constraints.
The current docs emphasize access and operational controls more than a standalone public pricing grid.
Endpoint rate limits
Create endpoints are limited to 60 requests per minute per organization, and read-style endpoints are limited to 600 requests per minute per organization.
Spend and tier controls
Anthropic also applies organization-level spend limits and tier-based rate limits, so app-level throughput planning cannot stop at the session API.
Runtime Envelope
Container and tool constraints affect architecture early.
These container and platform constraints shape what belongs in the first prototype.
| Dimension | Documented limit or behavior | Why it matters |
|---|---|---|
| Cloud container spec | Up to 8 GB memory, up to 10 GB disk, Ubuntu 22.04 LTS, amd64. | This is enough for many coding and data tasks, but it is still a bounded container, not an unbounded worker fleet. |
| Default network state | Network is disabled by default and must be enabled in the environment config. | A prototype that silently assumes internet access will fail until you configure the environment correctly. |
| Default tools | Built-in tools cover bash, file operations, web search, web fetch, glob, and grep. | You may not need extra infrastructure for the first version if the built-in toolset already matches your workload. |
Cost Planning
Think in workload cost, not in a simple seat-price mental model.
The docs do not present Managed Agents as a single fixed-price service. Plan around usage instead.
Watch token usage and session events so cost surprises show up in development instead of after launch.
Treat long-running sessions as a product cost decision, not just a prompt decision.
Keep the first environment small and explicit. Too much package or network scope makes debugging and security review harder.
Model selection still matters because Managed Agents is infrastructure, not a flat-priced bundle of work.
FAQ
Most planning mistakes come from assuming Managed Agents is a magic black box.
These are the most common expectation resets before a team moves from docs to budgeting.
Is there a simple public Managed Agents pricing table yet?+
Not in the current docs. What Anthropic documents clearly is beta access, per-organization endpoint limits, org-level spend controls, and the managed container envelope.
Can I rely on a documented default runtime environment?+
Yes. The container reference lists the current OS, memory, disk, language runtimes, package managers, and default network behavior.
Related Pages
Keep moving with the page that matches the next decision.
These pages are the shortest path to the next question most teams hit.
What is Claude Managed Agents?
Use this page when you need a compact explanation of the hosted runtime, the core concepts, and the workloads that match Managed Agents best.
Open pageClaude Managed Agents quickstart
Use this guide to build the first working Managed Agents session without mixing up agents, environments, sessions, and event streams.
Open pageClaude Managed Agents troubleshooting
Use this page when a session is failing, stuck, or acting unlike the runtime you thought you configured.
Open pageStay Updated
Want updates as Managed Agents evolves?
Use the waitlist if you want the next update without checking the docs every week.
Waitlist
Want updates as Managed Agents evolves?
Get a short email when this guide adds new examples, API changes, or managed-agents implementation notes.