DeepSeek V4 Flash Crashes Its Own Servers, and the Backup Plans Cost 10x More

DeepSeek V4 Flash's official release has already overwhelmed the company's own servers. The official API started returning 503 errors with a 'service too busy' message, and DeepSeek suggested users temporarily switch to other LLM API providers.

But moving to a third party won't necessarily be cheaper. DeepSeek's official cache-hit price is 0.02 yuan per million tokens; Alibaba Cloud's Bailian, Tencent Cloud, and ByteDance's Volcano Ark all charge 0.2 yuan — 10 times more.

Agents repeatedly read code and context, so cached tokens often make up the vast majority of usage. On a platform with a higher cache-hit price, your real costs will climb accordingly.