GLM 5.3 Flash needs about 189.9 GB of memory at Q4_K_M (a 188.7 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 5 quants, from 82.2 GB (IQ2_XXS) to 188.7 GB (Q4_K_M). Longer context needs more: at 32K tokens the KV cache alone adds about 3.0 GB.
GLM 5.3 Flash – 320 B parameters, 18 B active, multimodal, MIT licensed
GLM 5.3 Flash is a 320 B parameter model with 18 B active parameters. It supports multimodal inputs and tool calling. The model is quantized for llama.cpp and available in several quantization tiers.
Good at
- Multimodal inference
- Tool calling in XML‑ish format
- Large context up to 1 048 576 tokens
Limits
- Requires custom llama.cpp build for glm5next
- No GPU offload support yet
- Tool calls returned as empty array in OpenAI endpoint
Best for
- Local multimodal chat applications
- Tool‑enabled workflows on CPUs
- Research on sparse attention models
Who it’s for. Developers and researchers who need a large multimodal model that can run locally on CPU or GPU with a custom llama.cpp build. The model is suitable for those who can manage the memory requirements and are comfortable with the XML‑ish tool call format.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ2_XXS | 82.2 GB (2 parts) | 83.5 GB | Too big | Get file → | |
| Q2_K | 114.2 GB (3 parts) | 115.4 GB | Too big | Get file → | |
| IQ3_XXS | 120.7 GB (3 parts) | 122.0 GB | Too big | Get file → | |
| Q3_K_M | 149.1 GB (4 parts) | 150.3 GB | Too big | Get file → | |
| Q4_K_M | 188.7 GB (4 parts) | 189.9 GB | Too big | Get file → |
For 128 GB RAM, use Q2_K (114.16 GB file, ~115.4 GB memory). For 256 GB RAM, use Q4_K_M (188.67 GB file, ~189.9 GB memory). The IQ2_XXS tier (82.25 GB file) is suitable for 64 GB RAM but requires careful memory management.
What GPU do you need for GLM 5.3 Flash?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which GLM 5.3 Flash quant should you choose?
Q4_K_M is the usual starting point for GLM 5.3 Flash: it balances file size and quality well. Here it is a 188.7 GB file.
If memory is tight, the smallest file is IQ2_XXS at 82.2 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about GLM 5.3 Flash
How much VRAM does GLM 5.3 Flash need?
About 189.9 GB at Q4_K_M with an 8K context window. That is the 188.7 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run GLM 5.3 Flash on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 83.5 GB, so you would need a larger GPU or CPU offloading.
What is the file size of GLM 5.3 Flash GGUF?
It depends on the quant: IQ2_XXS is 82.2 GB, Q2_K is 114.2 GB, IQ3_XXS is 120.7 GB, Q3_K_M is 149.1 GB, Q4_K_M is 188.7 GB.
What context length does GLM 5.3 Flash support?
The GGUF header lists a trained context length of 1,048,576 tokens (1024K). The KV cache grows with the context you use: about 0.8 GB at 8K and 3.0 GB at 32K.
Where can I download GLM 5.3 Flash GGUF?
The files are on Hugging Face in the batiai/GLM-5.3-Flash-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does GLM 5.3 Flash use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run GLM 5.3 Flash on 8 GB VRAM?
No. The smallest quant requires 82.25 GB file size and ~83.5 GB memory at 8k context, far exceeding 8 GB VRAM.
Does the model support vision input?
Yes, it is a multimodal model, but vision support requires a custom llama.cpp build and is not available in standard builds.
Will tool calls be returned in JSON format?
No. The model emits tool calls in its own XML‑ish format; the OpenAI endpoint returns an empty tool_calls array.