DeepSeek V4 and Tencent's Huan Yuan new model are expected to launch in April.

According to Baijing Lab, both DeepSeek V4, led by Yao Shunyu, and the new Hunyuan model are expected to be released in April. The leak suggests that DeepSeek-V4, a multimodal large model refined by Liang Wenfeng for a long time, will not only see significant improvements in coding capabilities but also achieve breakthroughs in LTM (Long-Term Memory). Recently, OpenRouter launched two mysterious models: Healer Alpha and Hunter Alpha, which are speculated to be versions of DeepSeek V4; domestic IPs can directly access these large models. According to their self-descriptions, Hunter Alpha is a trillion-parameter model with a 1 million token context window, specifically designed for agent workflows, long-term complex tasks, and deep tool invocation. Healer Alpha is a full-modal interaction model with a 260,000-token context window, capable of natively perceiving visual and audio inputs, performing cross-modal reasoning, and executing complex multi-step tasks accurately and reliably. As of publication, neither model has been officially claimed by any party. Additionally, according to Baijing Lab, Yao Shunyu, Tencent’s Chief AI Scientist, will also release a new Hunyuan model (30B parameter class) in April. This will be his first major product since joining Tencent in December last year and is highly anticipated.