What GPU do I need to run zai-org/GLM-5.3-Flash?
321.3B parameters, published in F8_E4M3. View on Hugging Face
Set up GLM-5.3-FlashGLM-5.3-Flash is published by zai-org on Hugging Face, with 441,348 downloads and 1,848 likes to date. It's a Glm5NextForConditionalGeneration model built for image-text-to-text, published natively in F8_E4M3.
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision times a flat 1.2 overhead for activations and fragmentation. The KV-cache grows with context and is not in that factor; it is listed per model below where the architecture is published. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| FP8 (native) | 299.3 GB | 359.1 GB | RTX 4090 | 8 | $5.41/hr |
| INT4 (quantized) | 149.6 GB | 179.6 GB | RTX A5000 | 8 | $1.41/hr |
A GPU is only matched to a row if its hardware supports that precision, and the primary recommendation is always a single-GPU fit when one exists.
INT4 requires a quantized checkpoint actually published for this model, check its Hugging Face page before relying on this row.
Cheapest way to run GLM-5.3-Flash at its published (F8_E4M3) precision: 8× RTX 4090, at $0.677/hr per GPU ($5.41/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
GLM-5.3-Flash: KV cache by context length
The KV cache is the memory the attention layers hold for every token of context, on top of the weights. It is not part of the flat 1.2x overhead in the table above, grows with context length and with every concurrent request, and is why a long-context deployment needs more VRAM than the table shows.
KV-cache size for GLM-5.3-Flash is not published for this architecture, so no figure is shown.
GLM-5.3-Flash: common questions
Can GLM-5.3-Flash run on a single GPU?
No. At FP8 (native) it needs 359.1 GB of VRAM, more than a single 32 GB desktop card holds. The cheapest capable card in the live feed is a 48.0 GB RTX 4090, and it takes 8 of them.
Is GLM-5.3-Flash already quantized?
Yes. It is published in FP8, one byte per parameter, so the 359.1 GB figure is already a quantized footprint rather than a full-precision one. Only INT4 goes below it, at 179.6 GB. A GPU without FP8 tensor cores cannot run it as published, which is why cards here are matched on precision support and not on VRAM alone.
How many GPUs do I need to run GLM-5.3-Flash?
8 at FP8 (native). It needs 359.1 GB of VRAM and the cheapest capable live offer is a 48.0 GB RTX 4090, so 8 of them come to $5.41/hr in total.
Does quantizing GLM-5.3-Flash lower the GPU bill?
Yes. At FP8 (native) the cheapest live fit is 8 RTX 4090 cards at $5.41/hr. At INT4 (quantized) it drops to 8 RTX A5000 cards at $1.41/hr, provided a quantized checkpoint exists for it.
Weight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More GLM-5 models
- GLM-5.3-Flash-BF16 (321.3B, BF16)
- GLM-5.3-Flash-FP8 (321.3B, F8_E4M3)
- GLM-5.3-Flash-Uncensored-FP8 (321.3B, F8_E4M3)
- GLM-5.3-Flash-UNCENSORED-FP8 (321.3B, F8_E4M3)
- GLM-5.3-Flash (321.3B, BF16)
Lineage
Quantized and fine-tuned versions
Alternatives at this size
Other models for image-text-to-text within about a third of GLM-5.3-Flash's 321.3B parameters, from other model lines.
- Qwen3-VL-235B-A22B-Instruct (235.7B, BF16)
- MiniMax-M3 (427.0B, BF16)
- Inkling-Small (266.0B, BF16)
- step3 (321.0B, BF16)
- Intern-S1 (240.7B, BF16)
More on GLM-5.3-Flash
Fits on a 192 GB GPU at INT4: every model that fits in 192 GB.
Best vision-language models: how GLM-5.3-Flash ranks against the rest.
Related reading: RTX 4090 pricing and specs, How much VRAM you need for LLMs, Serving LLMs with vLLM, vLLM vs TensorRT-LLM vs SGLang, and Best GPU for LLM inference.
