GLM-5.2 Challenges Claude Opus with Huawei-Powered Performance
- GLM-5.2 scored a dominance rate of 74.4 on FrontierSWE, trailing Claude Opus by just 1% and surpassing GPT-5.5’s score of 72.6.
- The model was developed using Huawei Ascend chips, with no NVIDIA hardware involved, costing approximately $25 million in training expenses.
- Unsloth AI released quantizations reducing the model size from 1.51TB to a manageable 238GB while maintaining ~82% accuracy.
- Z.ai’s stock surged by 90% following the release of GLM-5.2, reaching a new all-time high.
GLM-5.2 is a significant advancement in AI models, offering competitive performance against established models like Claude Opus and GPT-5.5 without relying on NVIDIA technology.
The release has not only pushed Z.ai’s stock to new heights but also highlights the growing capabilities of Huawei-powered AI development in the global tech landscape.(Source)
