Companion

deepseek-chat — a retired name the API still resolves to DEEPSEEK_FLASH.

deepseek-flash — the general-purpose model: a 1M token context, up to 384K output tokens, and image input, which needs no separate vision model. Thinks unless Thinking disables it.

deepseek-reasoner — a retired name the API still resolves to DEEPSEEK_FLASH.

deepseek-v4-pro — the same limits and features as DEEPSEEK_FLASH, with the strongest capability and a lower concurrency limit.