I had the same query for both Deepseek v4 flash and kimi K2.5; "Orthotic sandals in germany"
[
](https://)
As you can see end to end time for Deepseek is almost double that of K2.5. Even though the token generation is double that of Kimi K2.5. So it seems like there is something else that makes Deepseek slow. I'm curious what it is.
I've also got to say Kimi K2.5 really sits in a sweetspot when it comes to making assistant searches. Not too long, and making good additional searches.
Textbox contain everything.
