DeepSeek's official pricing page now shows V4 Flash and V4 Pro side by side, each with three numbers: cache-hit input, cache-miss input, and output per 1M tokens.
Flash lands at $0.028 / $0.14 / $0.28 and Pro at $0.145 / $1.74 / $3.48. The 5× gap between cache-hit and cache-miss input is the detail worth engineering around: stable system prompts and tool schemas directly cut your input bill.
The same page warns that the older deepseek-chat and deepseek-reasoner aliases will eventually be deprecated — a warning that has since become a hard date: 2026-07-24.
Sources and publication record
Source material is linked for readers who want to verify the underlying announcement or documentation.