What hardware do you need to run Qwen3.5 122B-A10B?

Qwen3.5 122B-A10B at UD-Q4_K_M is 78 GB of weights, with a context ceiling of 256k tokens. Not yet rated: released after our last ratings pass. 10B active out of 125B — the large-memory option that still runs at a usable speed.

Cheapest machine that runs itStrix Halo Framework Desktop, 128GB at $3,449

Fastest of the ones listedMac Studio M5 Ultra, 256GB — 51 tok/s at 32k context

Honest answer on costPays back in 53 years at 500k tokens a day.

How good is it, really?

On the Artificial Analysis Intelligence Index v4.3 it scores 16 (reasoning mode; 18 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 $0.26 per million input tokens and $2.08 per million output (OpenRouter, cheapest active endpoint, checked 2026-09-15). 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, 128GB	$3,449	20 tok/s estimated	Pays back in 53 years	Run the numbers
DGX Spark GB10 Grace Blackwell, 128GB	$4,699	25 tok/s estimated	Pays back in 72 years	Run the numbers
Mac Studio M5 Max, 128GB	$5,099	26 tok/s estimated	Pays back in 78 years	Run the numbers
MacBook Pro M5 Max (16-inch), 128GB	$6,999	26 tok/s estimated	Pays back in 107 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: 125.1B, of which 10B are active per token
Quantisation: UD-Q4_K_M
Weights on disk: 78 GB
KV cache: 0.8 GB at 32k context — Hybrid: 12 of 48 layers hold a growing cache. Published only as an unsloth dynamic quant, which allocates bits differently from a plain Q4_K_M.
Maximum context: 256k tokens (256k)
Licence: Apache 2.0
Sources: source 1