China’s artificial intelligence landscape is set to witness significant advancements with the anticipated launch of DeepSeek V4 and a new Tencent Hunyuan model, both expected in April 2026.
These simultaneous releases indicate an acceleration in China’s AI development, focusing on practical applications over benchmark performance, according to Whale Lab. DeepSeek V4 is a multimodal large model capable of generating text, images, and video, aiming to improve coding capabilities and long-term memory, a challenge for large language models.
Earlier reports suggested a March release for DeepSeek V4, but Whale Lab indicates April 2026. A lighter variant, “V4 Lite,” appeared on DeepSeek’s website on March 9. Models “Healer Alpha” and “Hunter Alpha” on the OpenRouter platform could be early versions of DeepSeek V4, with Hunter Alpha being a trillion-parameter model and Healer Alpha a multimodal model with cross-modal reasoning.
DeepSeek has collaborated with Huawei and Cambricon to optimize V4 for domestic Chinese AI chips and partnered with Baidu to enhance its AI search functionalities. Tencent’s new Hunyuan model, led by Shunyu Yao, will have approximately 30 billion parameters and focus on in-context learning and agent usability.
Yao, a Tsinghua University graduate, is known for his work on AI agent frameworks and has advocated for a shift from benchmarking to scenario-driven applications. These releases follow other AI model launches in China, intensifying the industry race across multimodal capabilities, long-term memory, and real-world agent performance.
