Zhipu AI has officially launched GLM-4.7, a new model dedicated to autonomous programming that leverages "thought retention" technology to maintain reasoning logic over extended conversational contexts. This capability complements the "interleaved thinking" feature introduced in the earlier GLM-4.5 version, which enables the system to pause and deliberate before executing tasks. Compared to its predecessor, GLM-4.6, the latest iteration delivers a significant performance boost, achieving a 73.8% success rate on the SWE-bench Verified benchmark—a standard test for software engineering proficiency.
In addition to core coding capabilities, Zhipu AI highlights GLM-4.7’s strong performance in "aesthetic coding," demonstrating an ability to generate visually appealing websites and presentations. The company's blog showcases several fully functional websites allegedly created from single prompt inputs. Benchmark comparisons indicate that GLM-4.7 is now competitive with leading Western commercial models from providers like OpenAI and Anthropic, marking a tightening race in AI coding performance.
The model is accessible via the Z.ai platform and OpenRouter, and can also be downloaded locally through Hugging Face. It supports direct integration into development workflows such as Claude Code. Positioned as a cost-effective alternative, Z.ai claims that GLM-4.7 operates at just one-seventh the cost of comparable models while delivering competitive output quality.
