T
Thomas Mueller
We could not find documented latency or tool-calling figures per model, and we cannot see where the time of an answer goes. In our own test of 11 models the median reply time in the in-app chat ranged from 12 s to 42 s, with single replies up to 129 s.
Request:
  • a published benchmark per model for the website widget (latency and tool-calling reliability), including the basis for the "recommended" model
  • per answer, the time split into model, tools and verification in the conversation explanation
Impact: model choice based on data, and targeted tuning of tools.