GLM-5.3 is a 753B-parameter text Mixture-of-Experts model from Z.ai, built on the same base model as GLM-5.2 with every gain coming from post-training. It is aimed at complex coding and long-horizon agentic work, and reports a 50% improvement over GLM-5.2 on Z.ai's in-house Code Bench along with open-weights state of the art on Terminal Bench 3.0 and Agents' Last Exam. Post-training also produced a marked jump in cyber capability, where the model leads CyberGym for vulnerability discovery.
The model uses DeepSeek-style sparse attention (DSA) over a 1,048,576-token context, and supports reasoning, function/tool calling, and a multi-token-prediction (MTP) draft layer for speculative decoding.
This model is ready for commercial use.
This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party's requirements for this application and use case; see link to Non-NVIDIA GLM-5.3 Model Card
GOVERNING TERMS: Use of this trial service is governed by the NVIDIA API Trial Terms of Service. ADDITIONAL INFORMATION: Use of the model is governed by the GLM-5.3 License. The licence is MIT-like, with the addition that a Model-as-a-Service operator whose aggregate revenue exceeds 10 billion US dollars over any consecutive 12 months must pass a Z.ai security review before commercial use.
Global
Use Case: Agentic software engineering: repository-scale code generation and refactoring, terminal and tool-driven agents, long-horizon multi-step planning with failure recovery, and security research including vulnerability discovery. The 1M-token context also suits long-document and whole-codebase analysis.
Huggingface: 08/27/2026 via link
References:
Architecture Type: Transformer (sparse Mixture-of-Experts with sparse attention)
Network Architecture: GLM-5.3 (GlmMoeDsaForCausalLM) — 78 decoder layers (3 dense MLP + 75
MoE), 256 routed experts per MoE layer with top-8 routing plus 1 shared expert, DeepSeek Sparse
Attention with an indexer selecting 2,048 tokens per query, and 1 MTP layer.
Number of Model Parameters: 753B total, ~40B activated per token
Input Type(s): Text
Input Format(s): String
Input Parameters: One-Dimensional (1D)
Other Properties Related to Input: Context length up to 1,048,576 tokens. Thinking budget is
controlled by reasoning_effort, which accepts low, high, or max and defaults to max. In
the chat template clear_thinking defaults to false; chat scenarios should pass
clear_thinking=true explicitly.
Output Type(s): Text
Output Format: String
Output Parameters: One-Dimensional (1D)
Other Properties Related to Output: Reasoning content is returned separately from the answer.
Tool calls are emitted in OpenAI-compatible form.
Runtime Engines:
Supported Hardware:
Preferred 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.
GLM-5.3
Acceleration Engine: vLLM on NVIDIA Dynamo
Test Hardware: NVIDIA Blackwell (GB300)
This endpoint serves an NVFP4-quantized checkpoint on two-node GB300 workers (8 GPUs per worker), using data parallelism with expert parallelism across the 8 GPUs, an FP8 KV cache, CPU KV-cache offloading, and MTP speculative decoding.
GLM-5.3 may produce inaccurate, biased, or objectionable responses, and may err in multi-step reasoning, particularly in scenarios not well represented in its training data. This endpoint serves a 4-bit (NVFP4) quantized checkpoint, so outputs may differ from the full-precision model in edge cases. The model's cyber capabilities are a stated strength and warrant particular care: deployers should apply use-case-specific safety evaluation and appropriate guardrails.
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 model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.