DeepSeek's deepseek-chat and -reasoner endpoints are now retired
The cutoff passed Friday and legacy calls now error out. The drop-in is deepseek-v4-pro or -flash: 1M context, MIT open weights, and cheaper output.
Copy markdown
Your old model names now error out
deepseek-chat and deepseek-reasoner were fully retired after Jul 24, 15:59 UTC. Any app still hardcoding those names is failing right now, not next month.
The fix is a one-line swap
Keep your existing base_url and API key; just change the model to deepseek-v4-pro or deepseek-v4-flash. Both speak the OpenAI ChatCompletions and Anthropic APIs, so no client rewrite.
V4 is bigger, with a 1M context
V4-Pro is 1.6T params (49B active); V4-Flash is 284B (13B active). Both carry a 1M-token context and up to 384K output, with Thinking and Non-Thinking modes.
And cheaper than what it replaces
V4-Pro runs about $0.44 in / $0.87 out per million tokens; V4-Flash about $0.14 / $0.28. Cache hits cut input roughly 100x, so repeated-context agents get very cheap.
Or self-host and skip the treadmill
Both models ship as open weights on Hugging Face under MIT. Pull them once and no vendor cutoff can 404 your pipeline out from under you again.