What hardware do you need to run Qwen3 32B?

Qwen3 32B at Q8_0 is 35 GB of weights, with a context ceiling of 40k tokens. Near-lossless precision. Only worth it if you have the memory to spare; it halves your speed. On the cheapest machine that runs it, a Corsair AI Workstation 300, 64GB, it generates 4.4 tok/s at 32k of context, worked out from memory bandwidth.

How good is Qwen3 32B, really?

On the Artificial Analysis Intelligence Index v4.3 it scores 7 (reasoning mode), which puts it in the Below every hosted tier band. Fine for simple, well-specified tasks. Noticeably less capable than anything the big labs sell today. See the whole table.

• Summarising — good
• Translation — good
• Everyday coding — good
• Reasoning & maths — usable
• Agentic work — usable

Is Qwen3 32B still worth running?

Not on the two figures this site can compare. The intelligence index puts 21 of the current models here above Qwen3 32B, and not one of them asks more of a machine: at 32k of context it needs 43.4 GB and scores 7 (reasoning mode), where the strongest of the 21, Qwen3.8 27B, needs 18.6 GB and scores 34 (xhigh reasoning effort; 22 without reasoning).

The cheapest machine still sold that holds Qwen3 32B at 32k is the Corsair AI Workstation 300, 64GB at $1,700, which holds Qwen3.8 27B as well. This site quotes Qwen3.8 27B at 10 tok/s estimated there and Qwen3 32B at 4.4 tok/s estimated. Qwen3.8 27B itself runs on the Framework Desktop, 32GB at $1,269, $431 less.

What Qwen3 32B costs either way

Renting the same model costs $0.08 per million input tokens and $0.28 per million output (OpenRouter, cheapest active endpoint, checked 2026-10-29). 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. What a million tokens costs each way puts the two prices side by side.

Machines that run Qwen3 32B

Every machine here holds it to 32k, the longest setting below this model's own 40k limit, so the choice between them is speed and price rather than how much you can keep in the window.

The table is the cheapest machine in each family, not the only one that runs it: 28 of the 56 machines on this site hold it at 32k. Within a family only memory decides, so that is every Mac Studio, Strix Halo box, Mac mini and MacBook Pro with 64GB or more, every NVIDIA card with 96GB or more and the DGX Spark, 128GB. One table sets every machine against the others: price, the memory its GPU can use, what it holds and how long it takes to pay for itself.

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 longest context is the longest setting the calculator offers that the machine still holds this model at, cache included, and no machine is shown taking it past its own 40k limit. Each one opens the calculator on that machine at that length.

For scale, the calculator starts from 80 tok/s for a hosted API and times a local machine against it. Nothing above reaches it, and the quickest is 33 tok/s.

The cheapest machine that runs it is sold with 64 GB. There is a page on what 64 GB runs, machine by machine: it names the 9 machines here sold at that size, and the 32 current models the roomiest of them holds at 32k.

What has been measured running Qwen3 32B

Every speed in the table above is worked out from memory bandwidth, and that is not because nobody has run Qwen3 32B. One machine here has a published speed for it, and the table above leaves it out: it is no longer sold. It is below, as published and again at 32k of context, which is what every other speed on this page is taken at.

A published figure is what its source reports, at the window the model was run at. The 32k figure is that measurement read at 32k of context: every token generated reads the weights and the whole key-value cache, so the longer the window the slower it runs. A discontinued machine's price is the one it launched at. Each published figure links the source that took it.

Three more machines, at a shorter window

The table above lists the machines that run it at 32k, the context the calculator starts on, where Qwen3 32B needs 43 GB. Three more machines hold it at a shorter window, because the weights are the same size either way and the key-value cache is not. None of them is cheaper than the table above, which starts at $1,700.

One machine per family, cheapest first, from the same list as the table above. Each window is the longest setting the calculator offers that the machine still holds this model at, and it opens the calculator on that machine at that length. What it needs there is the weights plus the key-value cache at that window, measured against the memory each machine's GPU can address, which that machine's own page gives.

Qwen3 32B at Q4_K_M or Q8_0?

Qwen3 32B is one of two models this site lists at two precisions, and the heavier build is the subject of this page. The weights go from 19.8 GB at Q4_K_M to 34.8 GB at Q8_0. The key-value cache is the same size either way, 8.6 GB at 32k of context, because both builds share an architecture, so what a machine has to hold goes from 28.3 GB to 43.4 GB. 36 of the 56 machines on this site hold the lighter build at 32k, and 28 hold the heavier. The cheapest one still sold is the Radeon AI PRO R9700, 32GB at $1,299, card only, for the lighter build and the Corsair AI Workstation 300, 64GB at $1,700 for the heavier.

The heavier build costs speed as well, because every token generated reads the weights and the whole cache. On the one machine here where both builds have been measured, the Mac Studio M3 Ultra, 96GB, no longer sold, this site quotes 24 tok/s at Q4_K_M and 16 tok/s at Q8_0, both measured.

What the extra precision buys is the half this site has no figure for. The intelligence index has one score for Qwen3 32B, 7, which both builds here carry, and the five ratings above are the same five on both. The memory and the speed are the whole measured difference between them.

The specifics

Parameters

32.8B

Quantisation

Q8_0 — also listed here at Q4_K_M, which is 20 GB

Weights on disk

35 GB

KV cache

8.6 GB at 32k context. How weights and cache add up.

Maximum context

40k tokens (32k natively; 128k with YaRN)

Licence

Apache 2.0

Sources

bartowski/Qwen_Qwen3-32B-GGUF, Qwen/Qwen3-32B