Local AI Deals

Cheapest box to run Llama-4-Behemoth at Q8_0

Meta’s Llama-4-Behemoth is 2000B parameters, 288B 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

Llama-4-Behemoth at Q8_0 needs 2128GB.

Nothing on the Grid runs Llama-4-Behemoth at Q8_0 today. The largest Config, Mac Studio, M5 Ultra, 512GB, is 1624GB short.

Announced, weights not released. Parameter counts are Meta's; the K/V figure is an estimate from the Llama 4 family and is marked as one.

GeForce RTX 5090 is a card, not a computer: its price is the US street median plus $1,000 for the minimum host machine. NVIDIA's $1,999 MSRP is not a price anyone pays.

memory = 2000B × 1.063 B/param (8.5 bits) + 0.250 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (288B active × 1.063 B/param) = 306.0 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.