About running deepseek 4.1 flash

Lately I tested (via api) deepseek 4.1 flash. I like it quite a lot. What is the minimum budget needed to setup and run deepseek 4.1 flash at the same quality and token/s (~40) of most hosted solutions? I'd use it mostly for coding/reverse engineering/security analysis. So I need enough "juice" to support its full context. (Obviously with the same hw then I could try other models like mimo 2.6 or glm 5.3) A plus if the hw is self contained and won't need a full pc dedicated to it.