GLM-5 Model Released as 'AI Engineer'
A new open-source model, GLM-5, has been released with capabilities designed to autonomously build entire software systems. Unlike previous coding assistants, GLM-5 is engineered for multi-step planning and system-level reasoning, positioning it as an autonomous agent or "virtual team member." The release signals a shift where open models can be used for more complex engineering tasks like codebase scaffolding and automated feature delivery.
- The model was developed by Zhipu AI, a Chinese company spun out of Tsinghua University, which recently became the first publicly traded foundation model company through an IPO in Hong Kong. - GLM-5 is a Mixture-of-Experts (MoE) model with 744 billion total parameters, but only 40 billion are active during any given inference, making it more efficient than a dense model of equivalent size. It utilizes DeepSeek Sparse Attention to handle a long context window of over 200,000 tokens. - A significant geopolitical detail is that GLM-5 was trained entirely on Huawei Ascend chips, using the MindSpore framework, completely avoiding NVIDIA hardware. This is notable as Zhipu has been on the U.S. Entity List since January 2025, restricting its access to certain technologies. - On the SWE-bench benchmark for software engineering tasks, GLM-5 (at 77.8%) outperforms models like Gemini 3 Pro (76.2%) and GPT-5.2 (75.4%), though it trails behind Claude Opus 4.5 (80.9%). - The model is released under the permissive MIT License, allowing for unrestricted commercial use and adaptation, with its weights publicly available on Hugging Face. - In practical, real-world coding tests against newer models like Opus 4.6 and GPT-5.3 Codex, GLM-5 was significantly slower and required more steps to achieve a working result in a complex app-building task. - The development of autonomous AI agents is supported by a growing ecosystem of open-source frameworks like AutoGen, which focuses on multi-agent collaboration, and CrewAI, for orchestrating role-playing agents. - The shift towards "agentic engineering" aims to move beyond simple code completion to automating larger-scale software development tasks, a capability for which GLM-5 is explicitly designed.
