Apparently `deepseek-v4-flash` automatically routes to the new model, if you are using the official provider:

   The official release of the DeepSeek-V4-Flash API is now in public beta. The API calling method remains unchanged — simply set the model name to deepseek-v4-flash to use the latest version.

https://api-docs.deepseek.com/updates/#date-2026-07-31

I use DeepInfra. Their whole catalogue is openai compat. US based.