Massive Token Overestimation in Multiple Tool Calls due to Internal Unicode Escaping

I have also reproduced a similar issue with Chinese requests, and I am wondering if Chinese requests are within the scope of your current fix.