GLM-5.3-Flash NVFP4

Use it to run supported language models on your own computer for private chat, image-related tasks, or API-powered coding-agent workflows.

Availability

Setup script and model files linked from GitHub and Hugging Face

Pricing model

Project Maya is free

Deployment

Runs on the user's own hardware

Privacy

Nothing leaves your machine

Licence

NVIDIA Open Model Agreement governs download; base model is MIT-licensed

• The checkpoint was released on Hugging Face on 10/05/2026. ^[1]

  Dataset disclosures are published in the `nvidia/nemotron-post-training-v2` and subset of `nvidia/nemotron-post-training-v3` and subset of `nvidia/nemotron-pre-training-datasets` Hugging Face collections.

• The checkpoint is best suited for local AI hardware and uses approximately 166 GiB of vRAM. ^[1]

  This checkpoint is best suited for local AI hardware and uses approximately **166 GiB of vRAM**.

• The checkpoint requires provided community vLLM containers because upstream support for its NVFP4 scale-factor format has not yet landed. ^[1]

  **IMPORTANT**: the NVFP4 scale factors in this checkpoint are in a new lossless compression format, but upstream vLLM support for this format has not yet landed.

• The checkpoint is ready for commercial or non-commercial use under the stated NVIDIA Open Model Agreement download terms. ^[1]

  **GOVERNING DOWNLOAD TERMS:** Use of this model is governed by [NVIDIA Open Model Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-agreement/).

• The model is a quantized version of ZAI's GLM-5.3-Flash. ^{[1] [2]}

  The NVIDIA GLM-5.3-Flash NVFP4 model is the quantized version of ZAI's GLM-5.3-Flash model, which is an auto-regressive language model that uses an optimized transformer architecture.

• It accepts text, image, and video inputs and supports context lengths up to 1M. ^{[1] [2]}
• The checkpoint supports vLLM and SGLang. ^[2]

  To serve this checkpoint with [vLLM](https://github.com/vllm-project/vllm), start from `vllm/vllm-openai:glm53-flash-arm64-cu130` and run:

• The model is optimized for NVIDIA Blackwell hardware.
• NVFP4 quantization reduces disk size and GPU memory requirements by approximately 3.33x. ^[2]

  This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 3.33x.

• The listed Hugging Face release date is 09/09/2026. ^[2]
• Z.ai says GLM-5.3-Flash is available at one-tenth the price of GLM-5.2. ^[3]
• GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series. ^[3]

  We introduce GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series.

• The model has 320B total parameters and 18B active parameters. ^{[1] [3] [5]}

  With 320B total parameters and just 18B active parameters, it outperforms GLM-5.2 across benchmarks and real-world workloads at one-tenth the price, while approaching Claude Opus 4.8 on coding and agentic benchmarks.

• It supports controlling the thinking budget with the reasoning_effort parameter at low, high, and max levels. ^[3]

  - GLM-5.3-Flash supports controlling the thinking budget through the `reasoning_effort` parameter, which accepts three levels: `low`, `high`, and `max`.

• GLM-5.3-Flash API services are available on the Z.ai API Platform. ^[3]

  📍 Use GLM-5.3-Flash API services on <a href="https://docs.z.ai/guides/llm/glm-5.3-flash">Z.ai API Platform.

• The model supports local deployment with SGLang, vLLM, TokenSpeed, Transformers, KTransformers, and Unsloth. ^[3]
• Project Maya provides an OpenAI- and Anthropic-compatible API with tool calls. ^[4]

  - **Plugs into your tools:** OpenAI- and Anthropic-compatible API with tool calls, ready for coding agents.

• The application includes browser chat and a live monitor. ^[4]

  - **Everything in the box:** a chat and a live monitor in the browser, pictures, thinking levels, a one-command setup.

• Project Maya runs models using GPU, RAM, and NVMe storage on the user's machine. ^[4]
• The source says nothing leaves the user's machine. ^[4]

  Nothing leaves your machine.

• Project Maya is free. ^[4]

  <sub>Project Maya is free.

• The setup offers GLM-5.3-Flash, DeepSeek-V4.1-Flash-1017, and experimental MiniMax-M3.1-Flash support. ^[4]

  The setup also offers **[DeepSeek-V4.1-Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017)**