LLM Opex

Models

Qwen3.8-2.4T-A95B: GPU memory and hosting cost

Open weights · 2446.18 billion parameters · Mixture of experts · 262,144-token context · licence other

No cloud sells this model per token: to use it you run it yourself, on GPU instances you rent.

Run it yourself

It needs 6115.4 GB of GPU memory at bf16, the precision it ships in.

CloudCheapest instance that fitsGPUsAn hourA month, around the clock
OCI4 × BM.GPU.MI300X.8, joined by RDMA cluster network8 × MI300X 192 GB in each$192.00$140,160
Nebius3 × gpu-b300-sxm.8gpu-192vcpu-2768gb, joined by InfiniBand8 × B300 288 GB in each$228.00$166,440
AWS10 × p4de.24xlarge, joined by EFA8 × A100 80 GB in each$274.47$200,363
Azure4 × Standard_ND96isr_MI300X_v5, joined by InfiniBand8 × MI300X 192 GB in each$274.56$200,429
Alibaba Cloud8 × ecs.gn8v-8x.16xlarge, joined by eRDMA8 × GPU H 96 GB in each$467.93$341,587
Google Cloud6 × a3-ultragpu-8g, joined by GPUDirect RDMA8 × H200 141 GB in each$508.84$371,454

Several servers: the model is sized across identical servers that hold it together, on instance types whose vendor documents a network between servers; the price is all of them. It is shown where no one server holds the model, or where it costs less than the one that does.

More from Qwen

Similar size, open weights

List prices collected 21 November 2026, refreshed every three hours. No negotiated rates or taxes.