@square-soccer-55433 thanks for sharing that. A quick look shows that the Pinecone indexes themselves are returning your queries pretty quickly...most in sub-100ms. I do see a couple spikes up to 500-600ms, I think that's correlated with some of your queries using

include_values=true

which we know can increase latency and try to discourage for production workloads. Given how you described what you're experiencing, it seems like the latency may be coming from elswhere in the stack/application. - Do you have any instrumentation in your application that shows where those 30 seconds are being spent? - What language/library/framework are you using to manage the communication between your application, Pinecone and gpt-5? - Are you making the Pinecone calls in parallel or one-after-another? - How many results are you sending to gpt-5? Larger context windows will definitely start to slow down (as well as cost you a lot more) so it would be good to make sure you're sending just the minimal amount of context needed.