According to Deepseek.com news, a preview version of DeepSeek-V4 series have been released on April 4, including two strong Mixture-of-Experts (MoE) language models — DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens.
For further detail information, you could get the official Chinese document from DeepSeek-V4 预览版：迈入百万上下文普惠时代 (qq.com). Or you could get the English document from deepseek-ai/DeepSeek-V4-Pro · Hugging Face.