Alibaba has introduced 'Tongyi DeepResearch,' a new 30-billion-parameter agentic AI model that reportedly outperforms GPT-4 and DeepSeek-V3 in deep research capabilities. The model is noted for its efficiency, activating only 3.3 billion parameters at runtime and having been trained on just two NVIDIA H100 GPUs, marking a significant advancement in compute-efficient artificial intelligence. This new model's unique feature is its “agentic mid-training phase,” which allows it to learn reasoning, planning, and tool usage before the standard fine-tuning process. Alibaba has made Tongyi DeepResearch fully open-source, a move aimed at promoting greater accessibility and collaboration within the AI landscape. No specific market reaction or stock price movement directly linked to this announcement was available in the provided information.