Create vs read rate limits: Managed Agents 300/1,200, Sume plans

Yes, both keep create and read traffic in separate budgets. Anthropic's page lists Managed Agents create endpoints at 300 requests per minute and read endpoints at 1,200. Sume gives every key a per-minute budget by plan, with writes and reads counted separately and reads at forty times the write number.

Anthropic facts are from its Rate limits page (listed under Sources) and Sume facts from Authentication, read 2026-09-30.

What are the numbers on each side?

Anthropic's page lists the Managed Agents create and read rows as shown. Sume's plan table applies across all of /v1.

What counts as a read on Sume?

Any GET or HEAD, so polling status_url, events_url and result_url or listing runs. Two POSTs that submit nothing also count as reads: /v1/generation/admission-preview and the MCP endpoint itself. Creating runs, cancelling and uploads are writes.

How do I know which budget I hit?

A 429 names it in error.details.scope as read or write, and the ratelimit-* headers describe the budget the current request spent from. Back off on retry-after. An MCP tool call spends the write budget once for the run it creates, and a jobs_status poll over MCP spends none.

What are the limits of this comparison?

The Anthropic rows cover only Managed Agents and may change; the Sume table is the shipped default and ratelimit-limit on a response is the authority for your deployment. Unauthenticated Sume requests are limited per client IP, with a smaller read multiple.