Local AI Deals

Cheapest box to run Qwen3-Coder-480B-A35B-Instruct at BF16

Alibaba’s Qwen3-Coder-480B-A35B-Instruct is 480B parameters, 35B of them read per token. The number below is computed from that and each vendor’s published specs, not asserted. Switch the quant or the context and it recomputes in place.

Fit finderspecs, not prices
Quant
Context

Qwen3-Coder-480B-A35B-Instruct at BF16 needs 963GB.

Nothing on the Grid runs Qwen3-Coder-480B-A35B-Instruct at BF16 today. The largest Config, Mac Studio, M5 Ultra, 512GB, is 459GB short. At it fits with 266GB headroom.

GeForce RTX 5090 is a card, not a computer: every price shown for it is the cheapest listing we found plus $1,000 for the minimum host machine.

memory = 480B × 2.000 B/param (16 bits) + 0.254 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (35B active × 2.000 B/param) = 70.0 GB read per token
usable = memory − operating system reserve (8% on a Mac, 6% on a DGX Spark, 4% on a card)

Bandwidth is the vendor’s published figure. Every speed is an estimate and says so. Methodology.