DeepSeek Launches V4 Models Trained on Huawei Hardware, Reducing Reliance on Nvidia Chips

DeepSeek has unveiled its next-generation AI model, the open-source V4, which the Hangzhou-based startup said is competitive with leading closed-source models from U.S. Firms such as OpenAI and Google DeepMind. The company released two versions of the model on Friday, April 24, 2026. The V4-pro model features 1.6 trillion parameters, making it DeepSeek’s largest model to date by that metric, while the smaller V4-flash version contains 284 billion parameters. Both models support a context window of 1 million tokens, a significant expansion from DeepSeek’s previous flagship model, which had a context window of 128,000 tokens. DeepSeek stated that the V4 model is designed to work with agent frameworks including Claude Code and OpenClaw, reflecting an industry shift toward models capable of handling complex, multi-step tasks with reduced human input. The V4-pro version is positioned as a higher-end model with performance comparable to leading closed-source systems, particularly in agentic coding, world knowledge, STEM, and competitive programming. In maximum reasoning mode, DeepSeek said V4-pro outperforms all open-source models, though it still trails some frontier closed-source systems such as Google’s Gemini 3.1 Pro and OpenAI’s GPT-5.4 in certain areas, according to a paper released alongside the model. The V4-flash model delivers similar reasoning ability in some areas but operates faster and at lower cost than the Pro version, with comparatively weaker world knowledge and lower performance on more demanding agent-based tasks. Soon after the model’s release, Huawei announced “full support” for the V4 models using its range of Ascend chips and supernode systems for model inference. The Shenzhen-based tech giant provided further details about the collaboration during a live stream on Friday afternoon. Analysts from Huatai Securities noted in a client briefing that the V4 release explicitly mentions compatibility with domestic chips. Cambricon Technologies also quickly announced compatibility with DeepSeek’s new models following the announcement. DeepSeek emphasized that the V4 model is adapted to run on Huawei hardware, marking a step in China’s effort to build a self-sufficient AI ecosystem. The startup said the V4-Pro Max variant “redefines the state of the art for open models,” outperforming its predecessors in core tasks. The adaptation for Huawei hardware signals progress in domestic AI infrastructure and supports China’s broader drive for an independent AI sector.