Update: DeepSeek's deepseek-chat and -reasoner are now retired
The 15:59 UTC cutoff just passed: calls pinned to the old names now error. Repoint to deepseek-v4-flash or -pro — and mind the silent thinking-mode swap.
Copy markdown
The cutoff is now enforced
As of 15:59 UTC (7:59 PM GMT+4) today, DeepSeek stopped resolving the aliases deepseek-chat and deepseek-reasoner. A request still pinned to those names now returns an error instead of quietly routing to a model — so any integration you never migrated is breaking right now.
Repoint to V4 — flash or pro
Swap the model string to deepseek-v4-flash or deepseek-v4-pro; both speak the OpenAI ChatCompletions and Anthropic interfaces, so it is usually a one-line change. Flash runs $0.14 in / $0.28 out per 1M tokens (cache hits ~$0.003 in); Pro is $0.435 in / $0.87 out for frontier-class coding.
The trap: your reasoning can silently vanish
The old names were not simply renamed — deepseek-chat mapped to V4-Flash's non-thinking mode and deepseek-reasoner to its thinking mode. Rename blindly to deepseek-v4-flash and you drop reasoning; set the thinking parameter explicitly or watch answers quietly get worse.