GLM-5.2

Description

GLM-5.2 is a flagship long-horizon reasoning and coding model from Z.AI. It builds on GLM-5.1 with a solid 1M-token context window, stronger coding performance, flexible thinking effort levels, and architecture improvements for sparse attention and speculative decoding. The upstream GLM-5 GitHub model card lists GLM-5.2 as a 744B-parameter Mixture of Experts model with 40B active parameters.

GLM-5.2 is designed for long-horizon agentic work such as repository-scale coding, tool use, terminal operations, reasoning, and multi-step engineering tasks. The GLM-5.2 NVIDIA Inference Microservice (NIM) packages the model for NVIDIA GPU-accelerated deployment through NVIDIA GPU Cloud (NGC).

This model is ready for commercial or non-commercial use.

Third-Party Community Consideration:

This model is not owned or developed by NVIDIA. Please see link to Non-NVIDIA GLM-5.2 Model Card and GLM-5 GitHub repository.

License and Terms of Use:

Governing Terms: The NIM container is governed by the NVIDIA Software License Agreement and Product-Specific Terms for NVIDIA AI Products. Use of the model is governed by the NVIDIA Open Model Agreement. Additional Information: MIT License.

Deployment Geography:

Global

Use Case:

Use Case: GLM-5.2 is designed for complex long-horizon agentic engineering tasks, including advanced coding, tool use, terminal operations, repository-scale generation, multi-step reasoning, and sustained optimization across long context windows. It supports thinking effort controls that can balance response quality and latency.

Release Date:

NVIDIA GPU Cloud (NGC): 06/30/2026 via NVIDIA NGC GLM-5.2 model page

Reference(s):

References:

• GLM-5 GitHub repository
• GLM-5.2 model page
• Z.AI GLM-5.2 blog
• Z.AI GLM-5.2 developer documentation
• GLM-5 Technical Report (arXiv:2602.15763)

Model Architecture:

Architecture Type: Transformer
Network Architecture: Mixture of Experts (MoE) with DeepSeek Sparse Attention (DSA) and IndexShare sparse-attention optimization
Total Parameters: 744B (40B active)
Pre-training Tokens: 28.5T

GLM-5.2 uses a Mixture of Experts architecture for long-context text generation. The upstream configuration identifies the model architecture as GlmMoeDsaForCausalLM, model type glm_moe_dsa, 78 hidden layers, 256 routed experts, 1 shared expert, and 8 experts selected per token. GLM-5.2 supports a 1M-token context window through a maximum position embedding value of 1,048,576.

Languages: English, Chinese

Input:

Input Types: Text
Input Formats: String
Input Parameters: One-Dimensional (1D)
Other Input Properties: Supports multi-turn conversations with system prompts, user messages, and assistant responses. Supports thinking mode, tool calling, and a maximum context length of 1M tokens.

Output:

Output Types: Text
Output Format: String
Output Parameters: One-Dimensional (1D)
Other Output Properties: Supports text generation for reasoning, coding, and tool-oriented workflows. Output can include reasoning content when thinking is enabled.

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration:

Runtime Engines:

• SGLang

Supported Hardware:

• NVIDIA Blackwell: B200, B300, GB200, GB300
• NVIDIA Hopper: H20, H200

Operating Systems: Linux

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version(s)

GLM-5.2

Short Name: glm-5.2

Training, Testing, and Evaluation Datasets:

Training Dataset

Data Modality: Text
Text Training Data Size: [More than 10 Trillion Tokens]
Data Collection Method by dataset: Undisclosed
Labeling Method by dataset: Undisclosed
Properties: The upstream GLM-5 technical materials state that the GLM-5 series was pre-trained on 28.5T tokens. GLM-5.2 adds long-context and architecture improvements over GLM-5.1, including IndexShare sparse-attention optimization and improvements to the MTP layer for speculative decoding.

Testing Dataset

Data Collection Method by dataset: Undisclosed
Labeling Method by dataset: Undisclosed
Properties: Undisclosed

Evaluation Dataset

Evaluation Benchmark Score: Evaluated on benchmarks spanning reasoning, mathematics, coding, software engineering, terminal tasks, agentic tool use, and long-horizon workflows.
Data Collection Method by dataset: [Hybrid: Automated, Manually-Collected]
Labeling Method by dataset: [Hybrid: Automated, Manually-Labeled]
Properties: Upstream GLM-5.2 materials report improved long-horizon and coding performance over GLM-5.1, including 62.1 on SWE-bench Pro and 81.0 on Terminal-Bench 2.1 Terminus-2.

Evaluated on benchmarks from GLM-5 GitHub repository and GLM-5.2 model page:

Inference

Acceleration Engine: SGLang
Test Hardware: NVIDIA Blackwell (B200, B300, GB200, GB300), NVIDIA Hopper (H20, H200)

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal developer team to ensure this model meets requirements for the relevant industry and use case and address unforeseen product misuse.

Users are responsible for model inputs and outputs. Users are responsible for ensuring safe integration of this model, including implementing guardrails as well as other safety mechanisms, prior to deployment.

Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

Get Help

Enterprise Support

For help with NVIDIA NIM, customers with an active NVIDIA AI Enterprise subscription can file a support ticket.