Ailyre

Model release · DeepSeek

DeepSeek-V4-Pro

Model released

MIT open source; 1.6T/49B MoE; 1M context / max output 384K; Thinking/Non-Thinking; official price (per million tokens, CNY): off-peak input cache-miss ¥4.5 / cache-hit ¥0.15 / output ¥13.5, peak ×2; current official pricing page still lists deepseek-v4-pro independently (DeepSeek-V4-Pro-0813); the old “from 9/14 temporarily routed to V4.1-Flash” note no longer applies.

Availability

MIT; API deepseek-v4-pro; HF deepseek-ai/DeepSeek-V4-Pro

Source trace

Reviewed by Ailyre. Source evidence is presented separately from our summary.

api-docs.deepseek.com

api-docs.deepseek.com

No snapshot saved; read the original.