Best GPU for GLM-4.5 Air: VRAM Requirements
GLM-4.5 Air needs about 234.9 GB of GPU memory in BF16 and 60 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.
GLM-4.5 Air needs about 234.9 GB of VRAM in FP16, 118.3 GB in FP8 and 60 GB at 4 bits for inference with an 8,192-token context. In FP16 it fits on one NVIDIA B300 288GB; a 4-bit build fits on one NVIDIA A100 PCIe 80GB.
GLM-4.5 Air is a mixture-of-experts language model with 106B total and 12B active parameters per token from Z.ai, released in July 2025, with a 131,072-token context window and published under the MIT. Its weights alone take 212 GB in BF16 and 53 GB at 4 bits per parameter; the tables below add the KV cache and runtime memory, then list the GPUs that hold each setup.
Recommended GPUs for GLM-4.5 Air
Best for inference
1 GPU × NVIDIA H200 141GB
FP8 or INT8 weights, 118.3 GB at 8,192 tokens.
Best for fine-tuning
2 GPUs × NVIDIA B200 180GB
LoRA on BF16 weights, 275.6 GB.
Budget pick
1 GPU × NVIDIA RTX PRO 6000 Blackwell 96GB
Smallest card class that holds a 4-bit build, 60 GB.
GLM-4.5 Air VRAM requirements by precision
One request with an 8,192-token context, KV cache in BF16, plus a 10 % margin for activations and runtime buffers.
| Precision | Weights | KV cache | Total for inference |
|---|---|---|---|
| BF16 / FP16 | 212 GB | 1.5 GB | 234.9 GB |
| FP8 | 106 GB | 1.5 GB | 118.3 GB |
| INT4 | 53 GB | 1.5 GB | 60 GB |
GLM-4.5 Air VRAM by task
Fine-tuning rows assume a short sequence, batch 1 and gradient checkpointing; longer sequences add activation memory.
| Task | GPU memory |
|---|---|
| Inference at the full 131,072-token context (FP8) | 143.8 GB |
| QLoRA (4-bit base) | 68.9 GB |
| LoRA (BF16 base) | 275.6 GB |
| Full fine-tuning (BF16, AdamW) | 1,865.6 GB |
GPUs that fit GLM-4.5 Air
Number of cards whose combined memory holds each setup, assuming the model is split across them with tensor or pipeline parallelism. A dash means more than eight cards.
| GPU | Memory | BF16 inference (234.9 GB) | FP8 inference (118.3 GB) | INT4 inference (60 GB) | QLoRA (68.9 GB) | LoRA (275.6 GB) |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5080 16GB | 16 GB | — | 8 GPUs | 4 GPUs | 8 GPUs | — |
| NVIDIA GeForce RTX 4090 24GB | 24 GB | — | 8 GPUs | 4 GPUs | 4 GPUs | — |
| NVIDIA L4 24GB | 24 GB | — | 8 GPUs | 4 GPUs | 4 GPUs | — |
| NVIDIA GeForce RTX 5090 32GB | 32 GB | 8 GPUs | 4 GPUs | 2 GPUs | 4 GPUs | — |
| NVIDIA Tesla V100 32GB | 32 GB | 8 GPUs | 4 GPUs | 2 GPUs | 4 GPUs | — |
| NVIDIA A100 PCIe 40GB | 40 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA L40S 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA Quadro RTX 8000 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA RTX 6000 Ada 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA RTX A6000 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA A100 PCIe 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA A100 SXM 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H100 PCIe 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H100 SXM 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA RTX PRO 6000 Blackwell 96GB | 96 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H200 141GB | 141 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA B200 180GB | 180 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA B300 288GB | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA GB300 NVL | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
What is GLM-4.5 Air?
GLM-4.5 Air is the compact version of GLM-4.5 from Z.ai: a mixture-of-experts model with 106 billion total and 12 billion active parameters, with the same thinking and non-thinking modes for agents, coding and reasoning.
- Developer
- Z.ai
- Released
- July 2025
- Parameters
- 106B total, 12B active per token
- Context window
- 131,072 tokens
- License
- MIT
The published checkpoint also holds a multi-token prediction layer (about 110 billion parameters in all), not counted here. Z.ai also publishes FP8 weights.
Source: GLM-4.5 Air model card (as of November 17, 2026).
Frequently asked questions
- How much VRAM does GLM-4.5 Air need?
- For inference with an 8,192-token context and one request, GLM-4.5 Air needs about 234.9 GB in BF16, 118.3 GB in FP8, 60 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
- Can GLM-4.5 Air run on a single GPU?
- Yes. In BF16 it needs about 234.9 GB, which fits one NVIDIA B300 288GB (288 GB).
- How much GPU memory does fine-tuning GLM-4.5 Air take?
- QLoRA (4-bit base): about 68.9 GB; LoRA (BF16 base): about 275.6 GB; Full fine-tuning (BF16, AdamW): about 1,865.6 GB. Longer training sequences and larger batches add activation memory on top.
- What is the best GPU for GLM-4.5 Air?
- Best for inference: 1 GPU × NVIDIA H200 141GB (FP8 or INT8 weights, 118.3 GB at 8,192 tokens). Best for fine-tuning: 2 GPUs × NVIDIA B200 180GB (LoRA on BF16 weights, 275.6 GB). Budget pick: 1 GPU × NVIDIA RTX PRO 6000 Blackwell 96GB (Smallest card class that holds a 4-bit build, 60 GB). Check live offers before booking, as availability changes.
- Can GLM-4.5 Air run on an RTX 4090?
- Not on one card: even a 4-bit build needs about 60 GB, so it takes 4 RTX 4090 cards (96 GB combined) with the model split across them.
- Can GLM-4.5 Air run on an RTX 5090?
- Not on one card: even a 4-bit build needs about 60 GB, so it takes 2 RTX 5090 cards (64 GB combined) with the model split across them.
Estimates use the formulas of the LLM VRAM calculator (decimal GB). Real usage depends on the inference engine, quantization format and memory fragmentation: test on the target GPU before a long rental.
Explore the live market
Every offer on the live market comes from a reviewed supplier and is sold by EverXCloud Inc.