Deepseek v4 via deepseek themselves is significantly cheaper. Because (1) Huawei collab and (2) vLLM etc dont implement half of the inference optimisations dee

Deepseek v4 via deepseek themselves is significantly cheaper. Because (1) Huawei collab and (2) vLLM etc dont implement half of the inference optimisations deepseek proposed in their paper.