What hardware do you need to run Gemma 4 26B-A4B?

Gemma 4 26B-A4B at Q4_K_M is 17 GB of weights, with a context ceiling of 256k tokens. Not yet rated: released after our last ratings pass. 3.8B active out of 25B, so it is quick for its class.

Cheapest machine that runs itStrix Halo Framework Desktop, 32GB at $1,269

Fastest of the ones listedMac Studio M5 Ultra, 96GB — 105 tok/s at 32k context

Honest answer on costPays back in 160 years at 500k tokens a day.

How good is it, really?

On the Artificial Analysis Intelligence Index v4.3 it scores 17 (reasoning mode; 13 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.

Summarising — not rated
Translation — not rated
Everyday coding — not rated
Reasoning & maths — not rated
Agentic work — not rated

What it costs either way

Renting the same model costs $0.042 per million input tokens and $0.22 per million output (OpenRouter, cheapest active endpoint, checked 2026-09-09). 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

Machine	Price	Speed at 32k	Pay-back
Strix Halo Framework Desktop, 32GB	$1,269	41 tok/s estimated	Pays back in 160 years	Run the numbers
Mac mini M6, 32GB	$1,299	15 tok/s estimated	Pays back in 177 years	Run the numbers
MacBook Pro M5 (14-inch), 32GB	$2,399	13 tok/s estimated	Pays back in 339 years	Run the numbers
Mac Studio M5 Max, 36GB	$2,499	40 tok/s estimated	Pays back in 322 years	Run the numbers
DGX Spark GB10 Grace Blackwell, 128GB	$4,699	52 tok/s estimated	Pays back in 577 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: 25.2B, of which 3.8B are active per token
Quantisation: Q4_K_M
Weights on disk: 17 GB
KV cache: 0.9 GB at 32k context — 5 of 30 layers grow with the context, at two KV heads by 512 wide; the other 25 stop at a 1,024-token window.
Maximum context: 256k tokens (256k)
Licence: Apache 2.0
Sources: source 1, source 2