I’m sending a heavy prompt to Opus 4. Output tokens are 1500-1800.
I’ve had it work once today but subsequent tries are timing out with a 504 error. I can see it logging the request (and billing me for it) in the anthropic api logs so I know it’s making it that far. I suspect your timeouts are too short for this kind of heavy lift on Opus 4. When I switch the model it works again so I think you just need to raise this limit.