DeepSeek API Platform notes: V4 models and pricing

The API platform is platform.deepseek.com, with docs at api-docs.deepseek.com. Base URL in OpenAI shape: https://api.deepseek.com. Anthropic shape: https://api.deepseek.com/anthropic. Model names on the first-call page: deepseek-v4-flash, deepseek-v4-pro, deepseek-v4-flash-vision-exp.

I changed model strings in an existing OpenAI client and did not rewrite pagination. Thinking is enabled by default; effort maps low/high/max. The pricing table I opened lists peak and off-peak rates per million tokens, cache hit versus miss, and a 384K max output. I will not freeze those numbers here as if they cannot move — DeepSeek says they can.

See DeepSeek-V4-Pro, DeepSeek-V4-Flash, and agentic coding.