Sunk Cost sunkcost.ai Data checked 2026-09-03

What hardware do you need to run Gemma 4 31B it?

Gemma 4 31B it at Q4_K_M is 20 GB of weights, with a context ceiling of 256k tokens. Not yet rated: released after our last ratings pass. Sizes and prices are current.

Cheapest machine that runs itStrix Halo Framework Desktop, 64GB at $1,959
Shortest pay-backMac Studio M5 Max, 36GB — Pays back in 186 years
Fastest of the ones listedMac Studio M5 Ultra, 96GB — 35 tok/s at 32k context
Honest answer on costPays back in 203 years at 500k tokens a day.

How good is it, really?

On the Artificial Analysis Intelligence Index v4.3 it scores 15 (reasoning mode; 14 without), which puts it in the Haiku-class band. In the same band as Anthropic's cheap, fast tier. Every current OpenAI model scores above this band. Score source. See the whole table.

What it costs either way

Renting the same model costs $0.09 per million input tokens and $0.34 per million output (OpenRouter, cheapest active endpoint, checked 2026-09-03). Buying a machine only beats that if you use it hard enough, for long enough, that the hardware price divides down below the rental bill.

Machines that run it

MachinePriceSpeed at 32kPay-back
Strix Halo Framework Desktop, 64GB $1,959 7.4 tok/s estimated Pays back in 203 years Run the numbers
Mac mini M5 Pro, 48GB $2,299 8.9 tok/s estimated Pays back in 212 years Run the numbers
Mac Studio M5 Max, 36GB $2,499 13 tok/s estimated Pays back in 186 years Run the numbers
MacBook Pro M5 Pro (16-inch), 48GB $3,599 8.9 tok/s estimated Pays back in 332 years Run the numbers
DGX Spark GB10 Grace Blackwell, 128GB $4,699 7.9 tok/s estimated Pays back in 516 years Run the numbers

One machine per family, cheapest first. Speeds are measured where a public benchmark exists and estimated from memory bandwidth otherwise; the calculator says which for any configuration.

The specifics

Parameters
31.3B
Quantisation
Q4_K_M
Weights on disk
20 GB
KV cache
6.2 GB at 32k context — 5 of every 6 layers use a 1024-token sliding window.
Maximum context
256k tokens (256k)
Licence
Apache 2.0
Sources
source 1, source 2, source 3