Can GLM 4.5 Air run on the H200?
Yes. At 8K context the best fit is Q8_0, using about 112 GB of the 134 GB usable on the H200. Estimated speed ceiling: ~302 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | 43.6 GB Fits | 44.3 GB Fits | 48.6 GB Fits | 65.9 GB Fits | 782 |
| 3-bit (Q3_K_M) • | 54.7 GB Fits | 55.4 GB Fits | 59.7 GB Fits | 77.0 GB Fits | 619 |
| 4-bit (Q4_K_M) • | 69.4 GB Fits | 70.1 GB Fits | 74.4 GB Fits | 91.7 GB Fits | 486 |
| 5-bit (Q5_K_M) • | 79.2 GB Fits | 79.9 GB Fits | 84.2 GB Fits | 101 GB Fits | 424 |
| 6-bit (Q6_K) • | 93.6 GB Fits | 94.3 GB Fits | 98.6 GB Fits | 116 GB Fits | 358 |
| 8-bit (Q8_0) • | 111 GB Fits | 112 GB Fits | 116 GB Fits | 133 GB Tight | 302 |
| 16-bit (F16/BF16) • | 207 GB No | 208 GB No | 212 GB No | 230 GB No | — |
• real GGUF file size from unsloth/GLM-4.5-Air-GGUF. Speed is a bandwidth ceiling (4800 GB/s), not a benchmark. Methodology.
More
GLM 4.5 Air on every GPU and Mac · Every model that runs on the H200 · Other models that fit 64 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (zai-org/GLM-4.5-Air), NVIDIA specs. Data as of 2026-11-17.