News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
fnGLM-5

GLM-5

Open-weight model for engineering and agentic tasks

GLM-5 is a large language model released in 2026 by Z.ai. It targets complex systems engineering and long-horizon agent work. The model uses 744 billion parameters with 40 billion active per token. It aims to improve intelligence efficiency through scaling and asynchronous reinforcement learning.

Share on WhatsApp

  • 744 billion parameters with 40 billion active
  • Trained on 28.5 trillion tokens of data
  • Uses asynchronous RL infrastructure called slime
  • Supports context windows up to 202,752 tokens
  • Integrates DeepSeek Sparse Attention for efficiency

What it is

GLM-5 is a general-purpose language model designed for complex systems engineering and tasks requiring long-term planning. Developed by Z.ai, it represents a significant scale-up from its predecessor. The model contains 744 billion parameters, with 40 billion active during any given operation. It was trained on 28.5 trillion tokens of data. The developers state it achieves top performance among open-source models in reasoning, coding, and agentic capabilities.

How it came about

The model follows GLM-4.5 and GLM-4.7 in the GLM series. Z.ai launched GLM-5 in 2026 to address inefficiencies in training and deployment. The team focused on scaling parameters and pre-training data to improve artificial general intelligence efficiency. They also introduced a new infrastructure called slime to handle reinforcement learning at scale. This allowed for more detailed post-training iterations compared to earlier versions.

Inside the model

GLM-5 uses a mixture of experts architecture. It integrates DeepSeek Sparse Attention to lower deployment costs while keeping long-context abilities. The training process relied on slime, an asynchronous reinforcement learning system. This infrastructure improved throughput and efficiency during post-training. The model supports a context window of up to 202,752 tokens when using tools. It was built to bridge the gap between basic competence and excellence in pre-trained systems.

What it can do

The model performs strongly on academic benchmarks for reasoning and coding. On the AIME 2026 I test, it scored 92.7. It achieved 77.8 on SWE-bench Verified and 73.3 on the multilingual version. In terminal tasks, it scored 56.2 on Terminal-Bench 2.0. The developers claim it closes the gap with frontier closed models. It also handles browser-based tasks, scoring 62.0 on BrowseComp without context management.

How to use it

Z.ai offers GLM-5 through its API platform. The model weights are open and available for local deployment. Developers can use frameworks like vLLM, SGLang, and Transformers to run it. Specific versions of these tools are required, such as vLLM v0.19.0 or higher. The company provides recipes and cookbooks to help with setup on different hardware.

Questions people ask about GLM-5

What is the parameter count of GLM-5?

GLM-5 has a total of 744 billion parameters. However, it uses a mixture of experts design, so only 40 billion parameters are active for each token processed. This helps reduce computational costs during inference.

Can I run GLM-5 on my own hardware?

Yes, the model weights are open. You can deploy it locally using supported frameworks like vLLM, SGLang, KTransformers, or Transformers. Ensure you have the required software versions, such as vLLM v0.19.0 or newer.

What tasks is GLM-5 best suited for?

It is designed for complex systems engineering and long-horizon agentic tasks. It performs well in reasoning, coding, and using tools. Benchmarks show strong results in software engineering benchmarks and terminal-based interactions.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on GLM-5 yet, so the facts on this page come from the sources listed above: huggingface.co/zai-org/GLM-5. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.