Chinese AI firm Xhipu AI has recently unveiled a new open-source model called GLM 4.5, reinforcing it as a competitor to global players like OpenAI. The company, a spin-off from Tsinghua University, continues to expand its GLM (General Language Model) series, aiming to make advanced AI more accessible.
Among its recent releases, Zhipu AI has open-sourced versions of its GLM-4 series and GLM-Z1 inference models, including 9B and 32B parameter variants. These models, released under the permissive MIT license, support commercial use and offer developers greater freedom. Zhipu AI claims its GLM-Z1 inference model achieves notably fast inference speeds, up to eight times faster than some competitors, reaching 200 tokens per second on consumer-grade GPUs.
Among its recent releases, Zhipu AI has open-sourced versions of its GLM-4 series and GLM-Z1 inference models, including 9B and 32B parameter variants. These models, released under the permissive MIT license, support commercial use and offer developers greater freedom. Zhipu AI claims its GLM-Z1 inference model achieves notably fast inference speeds, up to eight times faster than some competitors, reaching 200 tokens per second on consumer-grade GPUs.
A key highlight is the GLM-Z1-Rumination-32B-0414, a “Rumination” model designed for autonomous AI agents. This model can actively search the internet, utilise tools, conduct in-depth analysis, and self-verify information to handle complex, open-ended queries. This represents a step towards more autonomous AI systems.
Additionally, Zhipu AI launched AutoGLM Rumination in March 2025, a free AI agent that combines research and task execution capabilities. Powered by GLM-Z1-Air and GLM-4-Air-0414 models, AutoGLM can perform deep research, web searches, travel planning, and report writing.
In October 2024, Zhipu also released GLM-4-Voice, an open-source end-to-end speech large language model. This model directly understands and generates Chinese and English speech, offering flexible adjustments for emotion, tone, speed, and dialect, and supports real-time interaction.
More recently, in July 2025, Zhipu AI launched GLM-4.1V-Thinking, an open-source multimodal reasoning model. This model is optimised for complex cognitive tasks, handling multimodal inputs like images, videos, and documents with improved performance, and runs on a single 3090 GPU. Zhipu states that GLM-4.1V-Thinking performs comparably to, and in some cases surpasses, larger models like OpenAI’s GPT-4o on various benchmarks.
Zhipu AI has also established an open-source fund for large language models, committing a thousand AI chip cards and 10 million yuan (USD 1.39 million) in cash to support open-source projects. This strategy aims to foster the broader AI industry ecosystem and accelerate technological advancements. The company has attracted significant investment from major Chinese tech firms like Alibaba Group Holding and Tencent Holdings, and has been actively involved in the “Special Migration Program” for OpenAI API users following OpenAI’s API block in some regions.