Where do you run Deepseek?

Do most people run DeepSeek through their API platform? I have tried experimenting with letting my agent route automatically to the most fitting LLM in terms of cost and what the "prompt" demands (use Opencode btw). Tested openrouter.ai routing, but it really didn't work. The closest I came was using standardcompute.com to automatically route between ds flash, pro and glm 5.2 on depending on task. I'm basically trying to reduce my costs (spend around $300 a month now). So wondering what other people are doing..