DeepSeek V4 arrives as a preview, with a million token context and Huawei chips in the picture
DeepSeek released preview versions of V4-Pro and V4-Flash on 24 April 2026 as open weights under the MIT licence, with 1 million tokens of context by default. Huawei said its whole Ascend supernode line supports them, a sign of a Chinese frontier model built around Chinese chips.