Chinese startup DeepSeek has unveiled a sneak peek of its latest V4 artificial intelligence model optimized for Huawei’s recently launched Ascend AI chips. The V4 model, available in two versions – Pro and Flash, boasts a 1-million-token context window. Pro aims for top-tier performance, competing with proprietary models, while Flash offers a more budget-friendly and quicker alternative.
DeepSeek highlighted that the V4 model is tailored to integrate seamlessly with agent frameworks like Claude Code and OpenClaw, aligning with the trend in the industry towards multi-step task execution by AI models. In its maximum reasoning mode, the Pro version surpasses all open-source rivals, as per a paper accompanying the preview. However, it still lags behind leading closed-source systems such as Google’s Gemini 3.1 Pro and OpenAI’s GPT-5.4 in certain aspects.
A notable development in DeepSeek’s latest release is the collaboration with Huawei hardware. According to a Reuters report, Huawei affirmed immediate support for V4 on its Ascend 950-based supernode clusters, with Flash’s training partially relying on Huawei chips. Huawei stated that the entire Ascend supernode product range now backs the DeepSeek-V4 series models after close technical cooperation. Notably, DeepSeek has not disclosed whether V4 was also trained on Nvidia chips, unlike its previous V3 and R1 models.
DeepSeek indicated that the Pro variant may be priced up to 12 times higher than Flash due to limitations in high-end computing resources, affecting its current availability. Reuters reported that DeepSeek anticipates a significant reduction in Pro pricing once Huawei’s Ascend 950 supernodes are widely deployed in the latter half of the year. The company faces computational constraints under US export regulations related to Nvidia chips and chip manufacturing equipment.