The tech circle at the end of the year has once again stirred up a storm. Zhipu Technology's latest GLM-4.7 model not only surpassed GPT-5.2 on the WebDev leaderboard of the Code Arena and ranked first among open-source large models, but also sparked heated discussions and testing enthusiasm among netizens. This domestically developed model, dubbed the "best alternative to Claude Code," has impressed people with its outstanding programming performance and flexible application capabilities.
During an AMA (Ask Me Anything) session, the Zhipu team revealed in detail the technical advancements of GLM-4.7. The team stated that the success of GLM-4.7 is attributed to optimization in the post-training phase, especially in supervised fine-tuning (SFT) and reinforcement learning (RL), where they adopted a more refined recipe, significantly improving the model's stability in practical applications. Additionally, the team emphasized that training and deployment costs were key considerations in the design, aiming to make the model run on consumer-grade GPUs while maintaining excellent performance.
GLM-4.7's application scenarios are diverse. Compared to previous versions, this model performs exceptionally well in multilingual coding, supporting multiple programming languages including Python and JavaScript, and shows greater flexibility in creative writing. Developers found that GLM-4.7 demonstrates higher understanding and logical abilities when debugging and executing complex tasks, and can even maintain role consistency in role-playing tasks, reducing the phenomenon of "breaking character."
Additionally, Zhipu announced the open-sourcing of its self-developed Slime framework during the AMA, aimed at improving the efficiency and stability of large model reinforcement learning. The team stated that the open-source ecosystem is their driving force for growth and will continue to uphold their open-source commitment, even after the company goes public, without reducing their investment in open source.
Through these technological innovations, Zhipu's GLM-4.7 is gradually demonstrating the competitiveness of domestic AI models on the international stage, earning high recognition from users. With the rapid development of AI technology, the Zhipu team also revealed that they will make more contributions to the widespread application of artificial intelligence next year, which is worth looking forward to.
