On the night of February 11, 2026, Zhipu unveiled and open-sourced its latest flagship model, GLM-5. This cutting-edge model has set a new benchmark in open-source Coding and Agent capabilities, boasting an expanded parameter scale of 744B (with 40B activated) and an increased pre-training dataset of 28.5T. It shines in complex system engineering and long-range Agent tasks, and has undergone deep inference adaptation with several domestic chips, such as Huawei's Ascend and Moore Threads. On the same day, DeepSeek carried out grayscale testing for its flagship model, extending the context window from 128K Tokens to a massive 1M Tokens and updating the knowledge base to May 2025. While the model prioritizes cost-efficiency, the grayscale iteration did not concurrently launch (Note: retained as is, meaning 'simultaneously launch') visual understanding or multimodal input functionalities. Moreover, Google's Gemini 3 Pro, which was released in November 2025, has been seamlessly integrated into Google Search and the Gemini App, offering robust support for long-text processing exceeding 2M and multimodal comprehension. Nevertheless, domestic users are required to access it through third-party platforms.