Can the GeForce RTX 5090 run GLM-4.5-Air?
Only with CPU offload at Q4_K_M.
GLM-4.5-Air is 106B parameters (MoE; all experts resident). At Q4_K_M the weights are about 59.2 GiB; with the 1.5 GiB runtime reserve that is 60.7 GiB against the card's 32GB, about 28.7 GiB more than the card has.
All three quants
32GB card| Quant | Weights | With reserve | Verdict | Headroom |
|---|---|---|---|---|
| Q4_K_Mthe everyday quant | 59.2 GiB | 60.7 GiB | Only with CPU offload | short 28.7 GiB |
| Q8_0near-lossless | 104.6 GiB | 106.1 GiB | Does not fit | short 74.1 GiB |
| FP16full weights | 197.4 GiB | 198.9 GiB | Does not fit | short 166.9 GiB |
Verdict rules: fits comfortably when weights plus reserve sit within 85% of VRAM; tight when they fit with little left for context; CPU offload when they exceed VRAM by up to 2x (runs, slowly, with layers in system RAM); no beyond that. The KV cache per token depends on the model's architecture and is not modelled here.
Not this card. These fit it comfortably at Q4.
Nothing in the consumer catalog fits GLM-4.5-Air at Q4 on a single card; this is multi-GPU or datacenter territory, or a job for hosted inference (see PicksByModel for per-token pricing).
What the GeForce RTX 5090 does run well → · GLM-4.5-Air on every card →
About the model
GLM-4.5-Air by Z.ai (Zhipu): 131,072-token native context, text, released 2025 under the MIT. Model card → · Hosted pricing on PicksByModel →