To be honest, connecting to the Hermes Model with GPT-5.4 always causes this kind of lag; the only good thing is that it's cheap. Sonnet 4.6 works perfectly fine, just expensive and not durable. Planning to use Sonnet for handling complex tasks and self-improvement, and switch GPT to monitoring.