Update: deepseek-chat and -reasoner go offline today; move to V4
Calls to those names start erroring today — repoint to deepseek-v4-flash or -pro. Also: PyPI now freezes new files on releases older than 14 days.
Copy markdown
The cutover is live
As of today, DeepSeek's deepseek-chat and deepseek-reasoner model IDs are discontinued, and calls to them now fail. Send non-thinking traffic to deepseek-v4-flash and thinking traffic to deepseek-v4-pro — both speak the OpenAI and Anthropic APIs, so it's a one-line model-name swap.
What V4 actually costs
V4 Flash runs about $0.14 per million input tokens and $0.28 output; V4 Pro is $0.44/$0.87 — both carry a 1M-token context and up to 384K output, in thinking and non-thinking modes. It stays one of the cheapest million-token APIs you can call.
PyPI freezes old releases
PyPI now rejects new file uploads to any release older than 14 days, closing a door on stolen publish tokens quietly poisoning long-stable versions. If you backfill platform wheels late, ship them within two weeks of cutting the release or the upload is refused — though PyPI's data shows only 56 of the top 15,000 projects ever did this.