抢跑GPT-5，智谱开源新SOTA模型，一句话搞出能看视频、发弹幕的B站！

Core Viewpoint - The release of GLM-4.5 marks a significant advancement in open-source large models, achieving state-of-the-art (SOTA) performance in various benchmarks and demonstrating a unique integration of capabilities [1][3][49]. Evaluation Metrics - The evaluation included 12 representative benchmarks such as MMLU Pro, AIME 24, and MATH 500, among others [4]. - GLM-4.5 ranked third globally in overall average scores, only behind closed-source models o3 and Grok4, while achieving first place in both open-source and domestic categories [5]. Model Architecture and Performance - GLM-4.5 utilizes a Mixture of Experts (MoE) architecture, featuring a total parameter count of 355 billion and 32 billion active parameters [9]. - The model boasts a generation speed of 100 tokens per second, significantly outperforming other AI models [6]. - Pricing for API calls is competitive, with input costs at 0.8 yuan per million tokens and output costs at 2 yuan per million tokens [8]. Practical Applications - GLM-4.5 can perform complex tasks such as coding and generating educational materials, showcasing its practical utility in real-world scenarios [21][25]. - The model has demonstrated superior performance in programming tasks compared to other open-source models, particularly in stability and task completion rates [24]. Training and Development - The training process involved multiple stages, starting with 15 terabytes of general pre-training data, followed by 7 terabytes of code and reasoning-related data [35]. - The model incorporates advanced techniques such as dynamic sampling temperature and adaptive pruning strategies to enhance stability and performance [48]. Community Engagement and Accessibility - GLM-4.5 is available for free public testing on platforms like chatglm.cn and Z.ai, promoting community engagement and feedback [12][50]. - The company has introduced a subscription model for developers, allowing unlimited access to GLM-4.5 for a nominal fee [55]. Conclusion - The launch of GLM-4.5 not only represents a technological leap for the company but also injects new vitality into the domestic open-source large model sector, showcasing China's capability to set new standards in AI [52][53].