SuperAPI AI Directory
Partner with us中文AI newsFavorites
DirectoryNews and tutorials

GLM-4.7 - Zhipu's flagship AI model, with further improved coding capabilities

Author: AI小集 · Original source: AI工具集

What is GLM-4.7

GLM-4.7 is an open-source model launched by Zhipu AI, focusing on improving coding capabilities, reasoning capabilities, and tool collaboration. The model performs excellently in multilingual programming, complex task planning, frontend design aesthetics, and more, and supports multiple programming frameworks, such asClaude Code, etc. In benchmark tests, GLM-4.7's coding capabilities reach the leading open-source level, and its reasoning capabilities are significantly enhanced. The model introduces interleaved, preserved, and turn-level thinking modes, making complex task execution more stable and controllable. The model now provides API services throughBigModeland inz.aithe full-stack development mode, the Skills module has been launched, providing developers with an efficient and intelligent programming experience.

GLM-4.7

Main features of GLM-4.7

  • Core coding capabilities: GLM-4.7 performs excellently in multilingual programming and terminal tasks, supports a "think first, then act" mode, and significantly improves the stability of complex tasks and code quality.
  • Frontend design optimization: Supports generating more modern and aesthetically pleasing web pages and slides, improves UI design quality, and reduces developers' time spent on style adjustments.
  • Tool calling capabilities: Performs excellently in tool calling and web browsing tasks, for example, scoring 87.4% on τ²-Bench and 67.5 points on BrowseComp, with greatly improved efficiency and accuracy.
  • Complex reasoning capabilities: Mathematical and reasoning capabilities are significantly enhanced, with a score of 42.8% on the HLE benchmark, an improvement of 12.4% over the previous generation, and it can handle complex logic and mathematical problems.
  • Thinking mode optimization: Introduces interleaved, preserved, and turn-level thinking modes, improving the execution stability and controllability of complex tasks, and is suitable for long-horizon tasks and multi-turn conversations.
  • Multimodal and full-stack development: Supports multimodal task collaboration and full-stack development, integrates the Skills module, and helps developers build applications with rich interaction and smooth experiences.

Performance of GLM-4.7

  • Coding capabilities (Core Coding)：
    • SWE-bench Verified: Score 73.8%, compared withGLM-4.6an increase of 5.8 percentage points, reaching the SOTA level of open-source models.
    • SWE-bench Multilingual: Score 66.7%, an increase of 12.9 percentage points over GLM-4.6, with significantly enhanced multilingual programming capabilities.
    • Terminal Bench 2.0: Score 41%, an increase of 16.5 percentage points over GLM-4.6, with greatly improved terminal task performance.
  • Tool using capabilities (Tool Using)：
    • τ²-Bench: Score 87.4%, an increase of 12.2 percentage points over GLM-4.6, with interactive tool calling capabilities reaching the open-source SOTA level.
    • BrowseComp (web browsing): Score 52.0%, an increase of 6.9 percentage points over GLM-4.6; in the BrowseComp test with context management enabled, the score was 67.5%, an increase of 10.0 percentage points over GLM-4.6, showing better performance in web browsing and toolchain management.
  • Complex reasoning capabilities (Complex Reasoning)：
    • HLE (Humanity's Last Exam): Score 42.8%, an increase of 12.4 percentage points over GLM-4.6, with significantly enhanced mathematical and reasoning capabilities.
    • MMLU-Pro: Score 84.3%, an increase of 1.1 percentage points over GLM-4.6, with stable performance in multi-domain reasoning capabilities.
    • GPQA-Diamond: Score 85.7%, an increase of 4.7 percentage points over GLM-4.6, with further improved reasoning accuracy.
GLM-4.7

Project address of GLM-4.7

  • Project official website：https://z.ai/blog/glm-4.7
  • GitHub repository：https://github.com/zai-org/GLM-4.5
  • HuggingFace model library：https://huggingface.co/zai-org/GLM-4.7

Application scenarios of GLM-4.7

  • Programming and Software Development: GLM-4.7 can generate high-quality multilingual code, serving as an intelligent programming assistant to improve development efficiency.
  • Front-End Development and Design: In web design and UI/UX design, it quickly generates modern and aesthetically pleasing layouts and color schemes, reducing the time front-end developers and designers spend on style adjustments.
  • Complex Task Planning and Execution: With interleaved thinking and turn-level thinking modes, GLM-4.7 can handle multi-step complex tasks, ensuring the accuracy and stability of long-horizon tasks.
  • Education and Learning: GLM-4.7 provides code examples and exercises for programming education, and also helps students improve their thinking skills through training on math and logic problems.
  • Office Automation: GLM-4.7 can automatically generate documents, reports, and data analysis code, reducing manual writing and formatting time, and improving office efficiency.