I’m not an infra person, so I might be missing something. I just want a coding agent running DeepSeek V4.1 Flash 1017 that’s cheap and behaves like the official API, not a slightly worse imitation.

Saw DeepSeek open-sourced DeepInfer and deepseek-recipe. My understanding: DeepInfer is the inference engine, deepseek-recipe keeps self-hosted output aligned with the official API. Has any coding agent vendor adopted either publicly? Or is this still DIY with Blackwell?

I don’t want to rent GPUs and debug CUDA. If no vendor has picked it up, just tell me to use the official API and wait.