On December 22, 2025, Z.ai (formerly Zhipu AI), a leading Chinese artificial intelligence company, announced the release of the latest version of its open-source large language model series , GLM-4.7 . This model represents a significant step forward in enhancing complex technical applications, particularly in programming and multi-step reasoning, with notable improvements over the previous version, GLM-4.6.
Key features: Advanced thinking techniques for better performance
GLM-4.7 features enhanced capabilities for handling advanced programming tasks, including multi-language code generation, executing operations requiring long chains of thought, and seamless integration with external tools in agenttic workflows. Key technical additions include:
- Interleaved Thinking : This allows the model to reflect and plan before each response or tool call, improving compliance with complex instructions and generation quality.
- Preserved Thinking : Maintains consistency of thinking across multiple rounds of dialogue, especially in long programming scenarios, to avoid rederiving reasoning from scratch.
- Turn-level thinking : provides flexibility in adjusting the depth of thinking according to the task, to balance speed and accuracy.
These features make the model ideal for modern software development, creating interactive content, controlling tools, and designing smart interfaces, with improvements in output quality such as producing elegant and consistent websites (Vibe Coding).
Performance and comparisons: A strong competitor to Western models
GLM-4.7 demonstrated superiority in multiple benchmarks, such as SWE-Bench (approximately 73.8%), AIME 2025 (95.7%), and HLE (42.8% with tools), outperforming models like Claude 4.5 Sonnet and GPT-5.1 in certain programming and mathematical tasks. It supports contextual code of up to 200,000–256,000 characters and output of up to 128,000 characters, with a MoE architecture for high computational efficiency.
The open-source model (MIT license) is offered on Hugging Face and ModelScope, with vLLM and SGLang support for local operation, and APIs via Z.ai, OpenRouter, and others, at a low cost (around $3 per month for the programming plan).
Competitive context: Strengthening Z.ai's position as "China's OpenAI"
The launch comes amid fierce competition in the open-source market, as Z.ai seeks to solidify its position as a leading developer of high-quality AI solutions, with a focus on programming communities. The company, which recently listed on the Hong Kong Stock Exchange (January 2026), is the first publicly traded firm to focus on core AGI models, reflecting its global ambitions despite geopolitical challenges.
In conclusion, GLM-4.7 represents a strategic evolution that combines innovation in smart thinking with integration with tools, opening up new horizons for customizable smart applications and enhancing competition in the open AI arena.
