Question: Undocumented 45 RPM Rate Limit for Claude Opus

[fleet walrus]

Hi team, we're consistently hitting a rate limit error on the anthropic/claude-opus-4 model, but we can't find any documentation about this specific limit.

The error states a limit of 45 RPM due to high demand:
Rate limit exceeded: limit_rpm/anthropic/claude-4-opus-20250522. High demand for anthropic/claude-opus-4 on OpenRouter - limited to 45 requests per minute. Please retry shortly.

Could you clarify if this is a temporary measure or a new permanent limit? We checked our account settings and the official docs but couldn't find this mentioned.

Thanks for your help!

@desert vigil

[desert vigil]

Are you specifically trying to use only Anthropic for Opus @fleet walrus ?

would recommend letting us route you to vertex at least

[fleet walrus]

Thanks for the suggestion! We need to stick with the native Anthropic model due to specific requirements.

My main question is about the 45 RPM rate limit itself. Could you clarify if this is a temporary measure, or a new standard we should plan for?

[desert vigil]

I removed the limit. we were recently working to secure more capacity so needed to have a limit to better spread our capacity.

Can i ask what the specific requirements are? the model is the same whether it’s hosted by Anthropic or by Google Vertex

Generally speaking, we’ve found Vertex to be a much much more stable provider for the Anthropic models? and they tend to have higher throughput (tokens per second)