GLM-5.3 Flash GPU requirements: 161 GB at 4-bit, 643 GB at BF16
GLM-5.3 Flash needs 643 GB of GPU memory for its weights at BF16, 328 GB at FP8 and 161 GB at 4-bit, plus 0.37 GB of KV cache for a 32K context. The cheapest node that runs it here is 4× A40 at 4-bit, for $0.24 an hour on spot.
Weights plus a 32K 16-bit KV cache, with 10% of the node's memory kept free. Billed per minute.
- Parameters
- 320B18B active per token
- Weights, 4-bit
- 161 GBBF16: 643 GB
- KV cache, 32K
- 0.37 GBOne sequence, 16-bit cache
- Context window
- 1MNative
How much VRAM GLM-5.3 Flash needs, format by format.
Weights are parameters × bytes per parameter, except the native FP8 row, which is the size of the published checkpoint. Each context column adds the KV cache of one sequence of that length.
| Format | Weights | + 8Kcontext | + 32Kcontext | + 128Kcontext | + 1Mcontext | Cheapest node, 32K |
|---|---|---|---|---|---|---|
| BF16Published weights, fine-tuning | 643 GB | 643 GB | 643 GB | 644 GB | 654 GB | 8× RTX PRO 6000 Blackwell · $1.52/h |
| FP8 (native)Hopper, Ada, Blackwell, MI300 | 328 GB | 328 GB | 329 GB | 330 GB | 340 GB | 8× L40 · $0.72/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 161 GB | 161 GB | 161 GB | 162 GB | 172 GB | 4× A40 · $0.24/h |
The serving engine adds its own buffers, and an 8-bit KV cache halves the context columns. We keep 10% of a node's memory free when we say a model fits. Speed is a separate question: it follows memory bandwidth and, for a mixture of experts, the active parameters.
GPUs that run GLM-5.3 Flash, cheapest first.
For each format, the smallest node of each GPU model that holds the weights and a 32K context, sorted by spot price. One row per GPU model; larger nodes of the same GPU work too.
| Format | GPU | GPU memory | Spot | On-demand | 730 h spot | Action |
|---|---|---|---|---|---|---|
| 4-bit | 4× A40Ampere · 161 GB needed | 192 GB | $0.24 | $0.36 | $175.20 | Deploy |
| 4-bit | 4× L40Ada Lovelace · 161 GB needed | 192 GB | $0.36 | $0.76 | $262.80 | Deploy |
| 4-bit | 4× RTX 6000 AdaAda Lovelace · 161 GB needed | 192 GB | $0.36 | $0.76 | $262.80 | Deploy |
| 4-bit | 4× RTX A6000Ampere · 161 GB needed | 192 GB | $0.36 | $0.68 | $262.80 | Deploy |
| FP8 (native) | 8× L40Ada Lovelace · 329 GB needed | 384 GB | $0.72 | $1.52 | $525.60 | Deploy |
| FP8 (native) | 4× RTX PRO 6000 BlackwellBlackwell · 329 GB needed | 384 GB | $0.76 | $1.56 | $554.80 | Deploy |
| FP8 (native) | 8× L40SAda Lovelace · 329 GB needed | 384 GB | $1.20 | $2.00 | $876.00 | Deploy |
| FP8 (native) | 4× H100 NVLHopper · 329 GB needed | 376 GB | $1.96 | $3.16 | $1,430.80 | Deploy |
| BF16 | 8× RTX PRO 6000 BlackwellBlackwell · 643 GB needed | 768 GB | $1.52 | $3.12 | $1,109.60 | Deploy |
| BF16 | 8× H100 NVLHopper · 643 GB needed | 752 GB | $3.92 | $6.32 | $2,861.60 | Deploy |
| BF16 | 4× MI300XCDNA 3 · 643 GB needed | 768 GB | $3.96 | $5.80 | $2,890.80 | Deploy |
| BF16 | 4× MI325XCDNA 3 · 643 GB needed | 1024 GB | $4.36 | $6.36 | $3,182.80 | Deploy |
Spot capacity can be reclaimed with a 2-minute notice; for a live endpoint, use on-demand. How spot works · serving a language model.
What a long context costs in memory.
Hybrid: 11 sparse latent-attention layers, the other 34 are linear-attention layers with a fixed-size state.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 0.09 GB | 0.74 GB |
| 32K tokens | 0.37 GB | 2.95 GB |
| 128K tokens | 1.48 GB | 11.8 GB |
| 1M tokens | 11.8 GB | 94.5 GB |
Serving several users
KV cacheEvery concurrent request keeps its own cache, so memory for context grows with users × context length. A server batching 8 requests of 32K needs eight times the single-sequence figure — 2.95 GB here.
- Recommended node4× A40
- Weights at 4-bit161 GB
- Context that fits, 1 sequence1M
Paged attention (vLLM, SGLang) allocates the cache in blocks as requests grow, so a node holds more short requests than the worst case suggests.
GLM-5.3 Flash, as published.
- Developer
- Z.aiHybrid linear-attention GLM, 1M-token context
- Parameters
- 320B total, 18B active per token
- Architecture
- Mixture of experts, 45 layersHybrid: 11 sparse latent-attention layers, the other 34 are linear-attention layers with a fixed-size state
- Context window
- 1,048,576 tokensAs set in the published configuration
- Published format
- FP8, 328 GB checkpoint
- Licence
- MIT
- Released
- 25 August 2026
- Source
- zai-org/GLM-5.3-Flashconfig.json read 26 October 2026
Serve it with vLLM
template: vLLMOn 4× A40, from the instance terminal:
$ vllm serve zai-org/GLM-5.3-Flash \ --tensor-parallel-size 4 \ --max-model-len 32768
Very recent architectures need a recent vLLM; the model card gives the minimum version and any extra flags.
GLM-5.3 Flash on a GPU — the questions that come up
How much VRAM does GLM-5.3 Flash need?
For the weights alone: 643 GB at BF16, 328 GB at FP8 and 161 GB at 4-bit. Add 0.37 GB of KV cache for one 32K-token context, or 11.8 GB at the full 1M window, and keep about 10% of the memory free for activations.
Can GLM-5.3 Flash run on a single GPU?
One RTX 4090 (24 GB): no. One RTX 5090 (32 GB): no. One H100 SXM (80 GB): no. One H200 SXM (141 GB): no. One B200 SXM (180 GB): yes, at 4-bit. Those answers assume a 32K context and the widest format that still fits.
What is the cheapest way to run GLM-5.3 Flash?
Here, 4× A40 at 4-bit: $0.24 per hour on spot and $0.36 on-demand, billed per minute — $175.20 for a 730-hour month on spot. That node has 192 GB of GPU memory for the 161 GB this setup needs.
Does GLM-5.3 Flash need less memory because only 18B parameters are active?
No. A mixture-of-experts model routes each token through a few experts, but every expert has to be resident in GPU memory, so the memory bill follows the 320B total. What the 18B active parameters change is the work per token: generation reads far less of the weights each step than a dense model of the same size would, which is why it is fast for its size.
Which format should I run it in?
GLM-5.3 Flash is published in FP8, a 328 GB checkpoint — the format its developers serve it in, so it is the safe default. BF16 keeps the published weights untouched and is the format for fine-tuning. FP8 halves the memory and runs in hardware on Hopper, Ada, Blackwell and MI300-class GPUs. 4-bit quantisation (AWQ, GPTQ, GGUF) quarters it, at a quality cost that is larger on small models than on big ones; check it against your own evaluation before relying on it.
How do I deploy it?
Pick the node in the table above and launch it with the vLLM template; the model downloads from Hugging Face (zai-org/GLM-5.3-Flash) on first start. Spread the weights over the node with --tensor-parallel-size equal to the GPU count, and set --max-model-len to the context you need. New architectures need a recent vLLM: check the model card for the minimum version.
Sizing another model? GPU requirements for 64 open models · serving a language model · fine-tuning.
GLM-5.3 Flash on 4× A40 for $0.24 an hour.
Launch the node with the vLLM template, point it at zai-org/GLM-5.3-Flash, stop it when you are done. Prepaid, billed by the minute.
Billed per minute from the moment the instance is reachable. Minimum credit $30, no subscription.