Core Insights - Zhipu AI is making significant strides towards becoming the first publicly listed company in the large model sector, recently unveiling its new generation model GLM-4.7, which has been open-sourced [2][36] Model Performance - GLM-4.7 focuses on enhancing programming and agentic tasks, showing improvements in reasoning capabilities [3] - The model has been tested across 17 benchmark tests, competing with models like GPT-5 and Claude Sonnet 4.5, achieving state-of-the-art results in two benchmarks: 95.7% accuracy in AIME 2025 and a score of 66.6 in BrowseComp-ZH [3][4] - In the MMLU-Pro benchmark, GLM-4.7 scored 84.3, while in GPQA-Diamond, it achieved 85.7, indicating competitive performance against other leading models [4] User Interface and Practical Applications - GLM-4.7 demonstrates significant improvements in UI quality, with enhanced capabilities in front-end development, allowing for the generation of HTML pages with clearer layouts and better color schemes [6][8] - The model's aesthetic capabilities have improved, particularly in 3D art displays and slide presentations, with a PPT 16:9 adaptation rate rising from 52% to 91% [16][21][22] - Users have reported positive experiences with the model's output quality, indicating a strong reception in practical applications [25] Reasoning Mechanism - GLM-4.7 introduces a new reasoning mechanism consisting of interleaved thinking, preserved thinking, and turn-level thinking, enhancing its ability to handle complex tasks [27][35] - Interleaved thinking allows the model to perform internal reasoning before generating responses, while preserved thinking retains information across multiple interactions, reducing inconsistencies [29][30] - Turn-level thinking enables control over reasoning processes within a session, optimizing performance based on task complexity [32] Company Overview and Financials - Zhipu AI is in the final stages of its IPO process, with a reported valuation of 24.377 billion yuan, marking a significant milestone for a Chinese large model company [39][40] - The company has shown rapid revenue growth, with projections indicating an increase from 60 million yuan in 2022 to 310 million yuan in 2024, and a substantial jump to 1.9 billion yuan in the first half of 2025 [41] - However, the company is also facing increasing losses, with adjusted net losses rising significantly due to high R&D expenditures, which reached 2.2 billion yuan [42][43]
“大模型第一股”打响上市前哨战!智谱GLM-4.7 刷新开源编程SOTA,修复代码、终端操作表现超Claude 4.5
AI前线·2025-12-23 07:29