CreateRun API: 10s latency when last user message is very short (but 2s when not short)

I noticed that some of my run creations were taking 10s. After some experimentation, it appears that when I the last user message is very short, creating a run takes 9 to 10s. But if the last user message is more substantive then creating a run only takes ~ 2 seconds. I was able to reproduce this over multiple tries.


If I add a user message that simply reads “Thanks” or “Got it”, the create run api call takes 10 seconds.

Instead, if I ask the assistant to read a document and provide a summary, run creation takes 2 seconds.

This is counterintuitive for two reasons: (1) I would expect a more complex user message to take more time to process and (2) the message isn’t a parameter of the create run api call, so not sure why it would have any effect at all on run creation speed.

Any one else seeing this behavior?