Zhipu announced at the Zhipu Open Platform Industry Ecology Conference held on July 2nd that it has received a joint strategic investment of 1 billion yuan from Pudong Venture Capital Group and Zhangjiang Group.
At the same time, Zhipu CEO Zhang Peng announced two latest achievements in Zhipu's collaboration with ecological partners towards AGI during the keynote speech: one is the open source release of the new generation of general visual language model GLM-4.1V-Thinking, which breaks through the performance limit of 10B level multimodal models with reasoning ability as the core; Secondly, MaaS has launched a new Agent aggregation platform called "Application Space", which fully activates AI capabilities in industry scenarios and collaborates with Z Fund to launch a special support plan for Agent Pioneers worth hundreds of millions of yuan.
Among them, the release and open source of the visual language model GLM-4.1V-Thinking marks a key leap from perception to cognition in the GLM series of visual models.
GLM-4.1V-Thinking is a general-purpose inference model that supports multimodal inputs such as images, videos, and documents, designed specifically for complex cognitive tasks. It introduces the "Chain of Thought Reasoning" mechanism based on the GLM-4V architecture and adopts the "Reinforcement Learning with Curriculum Sampling (RLCS)" strategy to systematically enhance the model's cross modal causal reasoning ability and stability.
It is worth noting that the lightweight GLM-4.1V-9B-Thinking model parameters are controlled at the 10B level, achieving performance breakthroughs while balancing deployment efficiency. In 28 authoritative evaluations such as MMStar, MMMU Pro, ChartQAPro, OSWorld, etc., this model achieved the best performance of a 10B level model in 23 of them, with 18 of them even equaling or surpassing the 72B parameter Qwen-2.5-VL, fully demonstrating the ultimate performance potential of small volume models.
At present, GLM-4.1V-9B-Thinking has been open sourced in sync with Hugging Face and the Magic community. It includes two models, namely the GLM-4.1V-9B-Base base model, hoping to help more researchers explore the capability boundaries of visual language models; GLM-4.1V-9B-Thinking， The model with deep thinking and reasoning ability, normal use and experience, are all this model.
When facing more complex real-world applications such as scientific research, education, autonomous agents, and multimodal assistants, GLM-4.1V-Thinking, which supplements reasoning abilities, truly solves open and information rich scenario problems in various industries, including long video understanding, image Q&A, disciplinary problem-solving, text recognition, document interpretation, Grounding, GUI Agent, and code generation.
In addition, Zhipu has launched a new ecological platform called "Agent Application Space" and launched the "Agents Pioneer Program", investing hundreds of millions of dollars to fully support AI Agent entrepreneurial teams.
Agent Application Space "is an AI agent capability aggregation platform aimed at enterprise customers and developers, excellent developers and enterprise users. The platform gathers rich Agent applications and Model Plug ins (MCP), providing out of the box, flexibly orchestrated component services and Agent applications, helping enterprises to access mature, secure, and controllable Agent capabilities at a low threshold without the need to build their own large model teams.
To some extent, the 'Agent Application Space' provides a one-stop development toolchain, a complete model calling interface, and a flexible application combination mechanism, which is an important starting point for upgrading the 'AI Primitives'. As Zhang Peng, CEO of Zhipu, said, "The future of AI is not only about competing model performance, but also about reconstructing production paradigms. In this era, every developer, designer, and entrepreneur is a co creator of the intelligent ecosystem. ”
In addition to open-source release of a new generation universal visual language model, release of a new Agents Pioneer program and application space, Zhipu's overseas business has also made significant progress.
The latest analysis report released by OpenAI recently pointed out that Zhipu has provided infrastructure solutions for governments and enterprises in Southeast Asian countries such as Vietnam, Indonesia, Malaysia, and Singapore, Middle Eastern countries such as the United Arab Emirates and Saudi Arabia, and African countries such as Kenya.
