DeepSeek Launches V4 AI Models, 1.6T Params

DeepSeek unveiled two new open‑source AI models, V4‑Pro (1.6 trillion parameters) and V4‑Flash (284 billion), each with 1 million‑token context window.
V4‑Pro matched OpenAI’s GPT‑5.4 on MMLU‑Pro benchmark, trailing only Google Gemini‑3.1‑Pro and Anthropic Claude Opus 4.6.
Training reportedly used Huawei’s Ascend 950 AI chips via Ascend supernode, after earlier U.S. accusations of using banned NVIDIA Blackwell chips.
The launch follows reports of Tencent and Alibaba discussing investment in DeepSeek at a valuation exceeding $20 billion.