DeepSeek V4 (1.6T params, 1M context) ships on Huawei chips with aggressive price cuts

DeepSeek released a preview of V4, its most powerful model to date — 1.6 trillion parameters, 1M-token context, trained on Huawei chips — and slashed pricing on the new flagship to undercut Western and Chinese rivals. Markets called the architectural advances 'predictable' and the launch failed to wow analysts.