Home / Will it run? / GLM-4.5-Air
Z.ai (Zhipu) · GLM-4.5
Will GLM-4.5-Air run on my GPU?
106B parameters (mixture of experts, 12B active per token, but every expert must sit in memory), 131,072-token context, MIT. At Q4_K_M the weights come to about 59.2 GiB; add 1.5 GiB for the runtime and you need roughly 60.7 GiB of VRAM before context.
Weights by quant
GiBQ4_K_M · the everyday quant59.2 GiB
Q8_0 · near-lossless104.6 GiB
FP16 · full weights197.4 GiB
Runtime reserve1.5 GiB
Card by card
0 of 18 fit at Q4| Card | VRAM | Q4_K_M | Q8_0 | FP16 | Headroom at Q4 | Buy |
|---|---|---|---|---|---|---|
| GeForce RTX 5090NVIDIA | 32GB | Offload | No | No | short by 28.7 GiB | $8,757 → |
| GeForce RTX 3090 (used)NVIDIA · used market | 24GB | No | No | No | short by 36.7 GiB | $2,404 → |
| GeForce RTX 4090 (used)NVIDIA · used market | 24GB | No | No | No | short by 36.7 GiB | $4,663 → |
| Radeon RX 7900 XTXAMD | 24GB | No | No | No | short by 36.7 GiB | $1,552 → |
| GeForce RTX 5060 Ti 16GBNVIDIA | 16GB | No | No | No | short by 44.7 GiB | $1,146 → |
| GeForce RTX 5070 TiNVIDIA | 16GB | No | No | No | short by 44.7 GiB | $1,792 → |
| GeForce RTX 5080NVIDIA | 16GB | No | No | No | short by 44.7 GiB | $2,399 → |
| Radeon RX 7800 XTAMD | 16GB | No | No | No | short by 44.7 GiB | $834 → |
| Radeon RX 9060 XT 16GBAMD | 16GB | No | No | No | short by 44.7 GiB | $812 → |
| Radeon RX 9070AMD | 16GB | No | No | No | short by 44.7 GiB | $1,052 → |
| Radeon RX 9070 XTAMD | 16GB | No | No | No | short by 44.7 GiB | $1,194 → |
| Arc B580Intel | 12GB | No | No | No | short by 48.7 GiB | $446 → |
| GeForce RTX 3060 12GB (used)NVIDIA · used market | 12GB | No | No | No | short by 48.7 GiB | $588 → |
| GeForce RTX 4070 SuperNVIDIA · used market | 12GB | No | No | No | short by 48.7 GiB | $1,156 → |
| GeForce RTX 5070NVIDIA | 12GB | No | No | No | short by 48.7 GiB | $1,086 → |
| Arc B570Intel | 10GB | No | No | No | short by 50.7 GiB | $299 → |
| GeForce RTX 5060NVIDIA | 8GB | No | No | No | short by 52.7 GiB | $550 → |
| GeForce RTX 5060 Ti 8GBNVIDIA | 8GB | No | No | No | short by 52.7 GiB | $809 → |
Headroom is what remains for the KV cache after weights and the 1.5 GiB reserve. How far it stretches depends on the architecture: grouped-query models are frugal, older dense models are not. If the table says tight, plan on a shorter context or a lower quant. Parameter count from the model's own card. The Buy column opens live Amazon listings; those links may earn us a commission.