GLM-4.7 Flash GPU requirements: 15.6 GB at 4-bit, 62.4 GB at BF16
GLM-4.7 Flash needs 62.4 GB of GPU memory for its weights at BF16, 31.2 GB at FP8 and 15.6 GB at 4-bit, plus 1.77 GB of KV cache for a 32K context. The cheapest node that runs it here is 1× A30 at 4-bit, for $0.05 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
- 30B3B active per token
- Weights, 4-bit
- 15.6 GBBF16: 62.4 GB
- KV cache, 32K
- 1.77 GBOne sequence, 16-bit cache
- Context window
- 198KNative
How much VRAM GLM-4.7 Flash needs, format by format.
Weights are parameters × bytes per parameter. Each context column adds the KV cache of one sequence of that length.
| Format | Weights | + 8Kcontext | + 32Kcontext | + 128Kcontext | + 198Kcontext | Cheapest node, 32K |
|---|---|---|---|---|---|---|
| BF16Published weights, fine-tuning | 62.4 GB | 62.9 GB | 64.2 GB | 69.5 GB | 73.4 GB | 1× A100 PCIe 80GB · $0.09/h |
| FP8Hopper, Ada, Blackwell, MI300 | 31.2 GB | 31.7 GB | 33.0 GB | 38.3 GB | 42.2 GB | 1× L40 · $0.09/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 15.6 GB | 16.1 GB | 17.4 GB | 22.7 GB | 26.6 GB | 1× A30 · $0.05/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-4.7 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 | 1× A30Ampere · 17.4 GB needed | 24 GB | $0.05 | $0.08 | $36.50 | Deploy |
| 4-bit | 1× L4Ada Lovelace · 17.4 GB needed | 24 GB | $0.05 | $0.08 | $36.50 | Deploy |
| 4-bit | 1× RTX 3090Ampere · 17.4 GB needed | 24 GB | $0.05 | $0.08 | $36.50 | Deploy |
| 4-bit | 1× RTX 4000 AdaAda Lovelace · 17.4 GB needed | 20 GB | $0.06 | $0.09 | $43.80 | Deploy |
| FP8 | 1× L40Ada Lovelace · 33.0 GB needed | 48 GB | $0.09 | $0.19 | $65.70 | Deploy |
| FP8 | 1× RTX 6000 AdaAda Lovelace · 33.0 GB needed | 48 GB | $0.09 | $0.19 | $65.70 | Deploy |
| FP8 | 2× L4Ada Lovelace · 33.0 GB needed | 48 GB | $0.10 | $0.16 | $73.00 | Deploy |
| FP8 | 2× RTX 4000 AdaAda Lovelace · 33.0 GB needed | 40 GB | $0.12 | $0.18 | $87.60 | Deploy |
| BF16 | 1× A100 PCIe 80GBAmpere · 64.2 GB needed | 80 GB | $0.09 | $0.19 | $65.70 | Deploy |
| BF16 | 2× A40Ampere · 64.2 GB needed | 96 GB | $0.12 | $0.18 | $87.60 | Deploy |
| BF16 | 2× A100 40GBAmpere · 64.2 GB needed | 80 GB | $0.18 | $0.34 | $131.40 | Deploy |
| BF16 | 2× L40Ada Lovelace · 64.2 GB needed | 96 GB | $0.18 | $0.38 | $131.40 | 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.
Multi-head latent attention: one compressed 576-value vector per token and layer, 47 layers.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 0.44 GB | 3.55 GB |
| 32K tokens | 1.77 GB | 14.2 GB |
| 128K tokens | 7.1 GB | 56.8 GB |
| 198K tokens | 11.0 GB | 87.8 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 — 14.2 GB here.
- Recommended node1× A30
- Weights at 4-bit15.6 GB
- Context that fits, 1 sequence108K
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-4.7 Flash, as published.
- Developer
- Z.aiSmall GLM mixture of experts (30B-A3B) for one GPU
- Parameters
- 30B total, 3B active per token31.2B in the published checkpoint (embedding tables, prediction or vision modules included); memory figures use this count
- Architecture
- Mixture of experts, 47 layersMulti-head latent attention: one compressed 576-value vector per token and layer, 47 layers
- Context window
- 202,752 tokensAs set in the published configuration
- Published format
- BF16, 62.4 GB
- Licence
- MIT
- Released
- 19 January 2026
- Source
- zai-org/GLM-4.7-Flashconfig.json read 26 October 2026
Serve it with vLLM
template: vLLMOn 1× A30, from the instance terminal:
$ vllm serve zai-org/GLM-4.7-Flash \ --tensor-parallel-size 1 \ --max-model-len 32768
Very recent architectures need a recent vLLM; the model card gives the minimum version and any extra flags.
GLM-4.7 Flash on a GPU — the questions that come up
How much VRAM does GLM-4.7 Flash need?
For the weights alone: 62.4 GB at BF16, 31.2 GB at FP8 and 15.6 GB at 4-bit. Add 1.77 GB of KV cache for one 32K-token context, or 11.0 GB at the full 198K window, and keep about 10% of the memory free for activations.
Can GLM-4.7 Flash run on a single GPU?
One RTX 4090 (24 GB): yes, at 4-bit. One RTX 5090 (32 GB): yes, at 4-bit. One H100 SXM (80 GB): yes, at BF16. One H200 SXM (141 GB): yes, at BF16. One B200 SXM (180 GB): yes, at BF16. Those answers assume a 32K context and the widest format that still fits.
What is the cheapest way to run GLM-4.7 Flash?
Here, 1× A30 at 4-bit: $0.05 per hour on spot and $0.08 on-demand, billed per minute — $36.50 for a 730-hour month on spot. That node has 24 GB of GPU memory for the 17.4 GB this setup needs.
Does GLM-4.7 Flash need less memory because only 3B 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 30B total. What the 3B 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?
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-4.7-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-4.7 Flash on 1× A30 for $0.05 an hour.
Launch the node with the vLLM template, point it at zai-org/GLM-4.7-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.