GLM-4.6 GPU requirements: 178 GB at 4-bit, 714 GB at BF16
GLM-4.6 needs 714 GB of GPU memory for its weights at BF16, 357 GB at FP8 and 178 GB at 4-bit, plus 12.3 GB of KV cache for a 32K context. The cheapest node that runs it here is 4× A100 PCIe 80GB at 4-bit, for $0.36 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
- 355B32B active per token
- Weights, 4-bit
- 178 GBBF16: 714 GB
- KV cache, 32K
- 12.3 GBOne sequence, 16-bit cache
- Context window
- 198KNative
How much VRAM GLM-4.6 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 | 714 GB | 717 GB | 726 GB | 763 GB | 790 GB | 4× MI325X · $4.36/h |
| FP8Hopper, Ada, Blackwell, MI300 | 357 GB | 360 GB | 369 GB | 406 GB | 433 GB | 8× RTX PRO 6000 Blackwell · $1.52/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 178 GB | 181 GB | 191 GB | 228 GB | 255 GB | 4× A100 PCIe 80GB · $0.36/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-4.6, 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× A100 PCIe 80GBAmpere · 191 GB needed | 320 GB | $0.36 | $0.76 | $262.80 | Deploy |
| 4-bit | 8× A40Ampere · 191 GB needed | 384 GB | $0.48 | $0.72 | $350.40 | Deploy |
| 4-bit | 8× A100 40GBAmpere · 191 GB needed | 320 GB | $0.72 | $1.36 | $525.60 | Deploy |
| 4-bit | 8× L40Ada Lovelace · 191 GB needed | 384 GB | $0.72 | $1.52 | $525.60 | Deploy |
| FP8 | 8× RTX PRO 6000 BlackwellBlackwell · 369 GB needed | 768 GB | $1.52 | $3.12 | $1,109.60 | Deploy |
| FP8 | 2× MI325XCDNA 3 · 369 GB needed | 512 GB | $2.18 | $3.18 | $1,591.40 | Deploy |
| FP8 | 4× H200 NVLHopper · 369 GB needed | 564 GB | $2.76 | $3.96 | $2,014.80 | Deploy |
| FP8 | 4× H200 SXMHopper · 369 GB needed | 564 GB | $3.00 | $4.36 | $2,190.00 | Deploy |
| BF16 | 4× MI325XCDNA 3 · 726 GB needed | 1024 GB | $4.36 | $6.36 | $3,182.80 | Deploy |
| BF16 | 8× H200 NVLHopper · 726 GB needed | 1128 GB | $5.52 | $7.92 | $4,029.60 | Deploy |
| BF16 | 8× H200 SXMHopper · 726 GB needed | 1128 GB | $6.00 | $8.72 | $4,380.00 | Deploy |
| BF16 | 8× MI300XCDNA 3 · 726 GB needed | 1536 GB | $7.92 | $11.60 | $5,781.60 | 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.
Grouped-query attention, 8 KV heads of 128 across 92 layers.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 3.09 GB | 24.7 GB |
| 32K tokens | 12.3 GB | 98.8 GB |
| 128K tokens | 49.4 GB | 395 GB |
| 198K tokens | 76.4 GB | 611 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 — 98.8 GB here.
- Recommended node4× A100 PCIe 80GB
- Weights at 4-bit178 GB
- Context that fits, 1 sequence198K
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-4.6, as published.
- Developer
- Z.aiAgentic and coding mixture of experts, 200K context
- Parameters
- 355B total, 32B active per token
- Architecture
- Mixture of experts, 92 layersGrouped-query attention, 8 KV heads of 128 across 92 layers
- Context window
- 202,752 tokensAs set in the published configuration
- Published format
- BF16, 714 GB
- Licence
- MIT
- Released
- 29 September 2025
- Source
- zai-org/GLM-4.6config.json read 26 October 2026
Serve it with vLLM
template: vLLMOn 4× A100 PCIe 80GB, from the instance terminal:
$ vllm serve zai-org/GLM-4.6 \ --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-4.6 on a GPU — the questions that come up
How much VRAM does GLM-4.6 need?
For the weights alone: 714 GB at BF16, 357 GB at FP8 and 178 GB at 4-bit. Add 12.3 GB of KV cache for one 32K-token context, or 76.4 GB at the full 198K window, and keep about 10% of the memory free for activations.
Can GLM-4.6 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): no. Those answers assume a 32K context and the widest format that still fits.
What is the cheapest way to run GLM-4.6?
Here, 4× A100 PCIe 80GB at 4-bit: $0.36 per hour on spot and $0.76 on-demand, billed per minute — $262.80 for a 730-hour month on spot. That node has 320 GB of GPU memory for the 191 GB this setup needs.
Does GLM-4.6 need less memory because only 32B 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 355B total. What the 32B 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-4.6) 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-4.6 on 4× A100 PCIe 80GB for $0.36 an hour.
Launch the node with the vLLM template, point it at zai-org/GLM-4.6, 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.