zai-org/GLM-4

GLM-4 series: Open Multilingual Multimodal Chat LMs | 开源多语言多模态对话模型

What it solves

GLM-4-0414 is a series of open-source large language models designed to provide high-performance capabilities in dialogue, reasoning, and complex task execution. It aims to bridge the gap between open-source models and proprietary systems like GPT-4o, particularly in areas like engineering code, function calling, and research-style writing.

How it works

The series utilizes different model sizes (9B and 32B) and specialized training paths:

  • GLM-4-32B-0414: Pre-trained on 15T of high-quality data and refined using human preference alignment, rejection sampling, and reinforcement learning to enhance instruction following and agentic capabilities.
  • GLM-Z1 (Reasoning): Developed via cold start and extended reinforcement learning focused on mathematics, code, and logic to enable "deep thinking."
  • GLM-Z1-Rumination: A specialized reasoning model trained through scaling end-to-end reinforcement learning, capable of longer thinking processes and using search tools to solve open-ended, complex problems.
  • YaRN: A context extension technique used to handle inputs exceeding the native 32K context length.

Who it’s for

  • Developers looking for high-performance open-source models for local deployment.
  • AI Researchers needing models with strong mathematical and logical reasoning capabilities.
  • Software Engineers requiring a model capable of generating engineering-grade code and performing function calling for agent tasks.
  • Users in resource-constrained environments who can utilize the lightweight 9B models.

Highlights

  • Diverse Model Range: Offers Base, Chat, Reasoning, and Rumination variants across 9B and 32B parameter sizes.
  • Strong Agent Capabilities: High performance in function calling, Artifact generation, and search-based Q&A.
  • Deep Reasoning: The Z1 series provides advanced capabilities for complex mathematical and logical tasks.
  • Broad Integration: Officially implemented in vLLM, transformers, and llama.cpp.

Related

  • Project
    zai-org/GLM-V

    A series of vision-language models (GLM-4.6V, 4.5V, 4.1V) that enable complex multimodal reasoning, native function calling, and precise visual grounding for agents.

  • Dispatch
    Open-source LLMs as LangChain Agents

    Hugging Face demonstrates that open-source LLMs, specifically Mixtral-8x7B, are now capable of powering agent workflows and can outperform GPT-3.5 in general-purpose reasoning tasks.

  • Dispatch
    Gemma 4 Release Notes: High-Intelligence Open Models for Reasoning and Agents

    Google DeepMind has released Gemma 4, a family of open models under an Apache 2.0 license designed for advanced reasoning, agentic workflows, and efficient on-device deployment.

  • Project
    zai-org/GLM-5

    GLM‑5 is an open‑source series of very large LLMs (320 B‑744 B total parameters) optimized for multimodal input, 1 M‑token contexts, and long‑horizon agentic tasks. It introduces hybrid sparse/linear attention, IndexShare, and a custom RL engine (slime) to achieve frontier‑level performance on coding and planning benchmarks while offering cheap FP8 inference. Models are downloadable from Hugging Face/ModelScope and can be served with SGLang, vLLM, KTransformers, etc., with optional ‘reasoning_effort’ control.

  • Dispatch
    Llama 3.1 Release Notes: Multilinguality, Long Context, and 405B Model

    Meta has released Llama 3.1, featuring models in 8B, 70B, and 405B sizes with 128K context length, multilingual support for 8 languages, and a permissive license allowing synthetic data generation.