Ive been building LLM wrappers and small integrations for a couple years now, mostly sticking to the big players like OpenAI or Anthropic since their infra is pretty much set it and forget it. But lately Ive been really digging DeepSeek-V3 for a coding assistant tool Im building for some local devs here in Berlin. The model performance is insane for the price, but Im hitting some weird wall with latency when Im calling their API directly from my current Vercel setup. It feels like the handshake takes forever or maybe its just a routing thing since their servers are mostly in Asia?

Ive got about $100 a month earmarked for the hosting side of this project and I need something that wont add 200ms of overhead just to get the request out the door. I tried a simple proxy on AWS but it didnt really help much and honestly Im getting tired of tweaking VPC settings for a side project that needs to be live by next Friday. Is there a specific provider that plays nice with DeepSeeks endpoints or maybe a middleware layer that handles the edge caching better? What are you guys using to host your backend when youre hitting DeepSeek heavily?