Zhipu.AI, one of China´s leading artificial intelligence companies, has announced the open-sourcing of its advanced General Language Models (GLM) series, including the GLM-4 and GLM-Z1 inference models. These new releases include the GLM-Z1-32B-0414 model, which achieves a substantial 8x speed advantage over DeepSeek-R1, delivering up to 200 tokens per second on consumer GPUs. Such accelerated inference speeds, made possible through innovations in GQA optimization, quantization, and speculative sampling, provide Zhipu with a decisive edge in Artificial Intelligence efficiency as it expands its footprint globally.
The company also introduced the GLM-Z1-Rumination-32B-0414 model, which marks a shift toward more autonomous and proactive Artificial Intelligence agents. This model demonstrates enhanced capabilities: actively searching the internet, utilizing available tools, self-verifying responses, and answering complex, open-ended queries. The expanded model suite also includes the core GLM-4-32B-0414, tailored for agent-like behavior with real-time code generation across formats including HTML, CSS, JS, and SVG, directly enhancing developer productivity. Smaller 9B-parameter variants of both GLM-4 and GLM-Z1 have been released to accommodate the needs of resource-constrained environments, all under the open and permissive MIT license.
Complementing this open-source strategy is the launch of Z.ai, Zhipu´s new global-facing platform, designed to widen access to their Artificial Intelligence models. Z.ai provides a web-based interface and dedicated app for users around the world to interact directly with these latest models. At the enterprise level, Zhipu continues to grow its Model-as-a-Service (MaaS) platform, offering robust API access to a range of GLM-powered solutions, accommodating both commercial and free usage tiers. As Zhipu.AI accelerates global expansion through technical innovation and ecosystem openness, industry observers view these moves as indicative of its readiness for a potential IPO and a broader role in shaping the future of open-source Artificial Intelligence worldwide.
