Core Insights - The article highlights the official launch and open-sourcing of the GLM-4.7 model by Zhipu AI, which is optimized for programming scenarios [1] Group 1: Model Performance - The GLM-4.7 model has achieved leading scores in various programming benchmark tests, including 73.8% in the SWE-bench-Verified test [1] - In the LiveCodeBench V6, the model reached a score of 84.9% [1] - The model scored 87.4 points in the tool invocation evaluation τ-Bench [1] Group 2: Features and Enhancements - The model supports a retention-based thinking mechanism for optimizing performance in long-range tasks [1] - There is an improvement in the visual aesthetics of front-end code and document generation, with a notable increase in the 16:9 adaptation rate for PPTs [1] - The model is currently available through API services provided by BigModel.cn [1]
智谱宣布GLM-4.7上线并开源：编程、推理与智能体三个维度实现突破
Feng Huang Wang·2025-12-23 02:24