The China International Capital Corporation Limited (CICC) took stock of the year 2025, noting that the global technology of large-scale models continued to advance, achieving remarkable progress in their application to productivity scenarios. Enhancements were observed in reasoning, programming, agentic capabilities, and multimodal functionalities. Nevertheless, challenges persist in terms of the stability and hallucination rate within their general capabilities. Looking forward to 2026, CICC anticipates that large-scale models will make further breakthroughs in areas such as reinforcement learning, model memory, and context engineering. These advancements are expected to facilitate a shift from short-context generation to tasks involving long chains of thought, and from text-based interactions to native multimodal interactions, thereby taking another stride toward the long-term goal of Artificial General Intelligence (AGI).