Zhipu AI Releases GLM-5, 745B Open Frontier Model Trained on Domestic Chips
On February 11-12, 2026, Zhipu AI (Z.ai) released GLM-5, a ~745 billion parameter open-weight frontier model. Trained entirely on Huawei Ascend chips with no NVIDIA dependency, it achieves competitive results on coding and agentic benchmarks, tops several open-source leaderboards, and triggered a significant stock surge for the publicly listed company.
TLDR
Zhipu AI released GLM-5 around February 11-12, 2026. The large-scale open-weight model (reported ~744-754B parameters, MoE architecture in some descriptions) posts strong scores on coding and reasoning benchmarks, approaching or matching aspects of leading proprietary models while running at substantially lower cost. It was trained exclusively on domestic Huawei Ascend hardware using MindSpore, marking a notable demonstration of China’s self-reliant AI stack under US export controls. The company’s Hong Kong-listed shares rose sharply on the announcement.
Technical and Geopolitical Significance
GLM-5 ships under open licensing (MIT-style references in coverage) with weights available via Hugging Face and Chinese platforms. It excels in coding and long-running agent tasks per contemporaneous reports and leaderboards. The complete avoidance of restricted NVIDIA hardware is presented as both a technical and strategic milestone.
Market reaction was immediate: the listed entity saw double-digit percentage gains in a single session. Demand reportedly overwhelmed initial capacity, leading to pricing adjustments in some channels.
Why this story matters
GLM-5 provides concrete evidence that frontier-class performance is achievable outside the dominant US GPU ecosystem. For developers and organizations prioritizing cost, sovereignty, or avoiding export-controlled supply chains, domestic Chinese models at this scale change the menu of practical options. The open release also expands the global pool of high-capability base models available for fine-tuning and research, while raising parallel questions about evaluation transparency, safety practices, and the speed of capability diffusion.
Sources
- Company and Reuters-linked reporting on the February 11, 2026 release of GLM-5 (Zhipu / Z.ai), parameter count, training stack (Huawei Ascend + MindSpore), and benchmark claims.
- Coverage of stock reaction and demand surge (28-34% gains cited across reports).
- Hugging Face and OpenRouter references confirming availability and relative ranking in the February 2026 timeframe.
- Technical summaries noting competitive coding/agentic performance versus Claude/GPT/Gemini families at the time.
Featured Image Alt Text
Zhipu GLM-5 release visual with large parameter scale, open model, and Huawei Ascend chip references
Tags
GLM-5, Zhipu AI, Z.ai, Open Weights, Frontier Model, Domestic Chips, China AI, Cost Efficient