What can you run with 32 GB of memory?

A model does not get the 32 GB. On the Mac mini M6, 32GB it gets 21 GB and on the Radeon AI PRO R9700, 32GB it gets 31 GB, because the system keeps a share of one and the card keeps a margin free on the other. That is 23 of the 51 current models at 32k of context on the first and 35 on the second, and the strongest of them is Qwen3.8 27B.

Run the numbers on Qwen3.8 27B at 32 GB

Jump to: The machines sold with 32 GB · Models that fit in 32 GB · What 36 GB adds over 32 GB · How much context 32 GB leaves room for · Does a 32 GB machine pay for itself?

The machines sold with 32 GB

There are six machines here with 32 GB, and what they hand a model is not the same figure. The fourth column is what each one holds at 32k of context, and it is the number that machine's own page prints.

List prices, and the memory each maker publishes. A machine marked previous is one that is no longer sold, priced at what it launched at. Graphics cards are priced as the card alone, so add the PC around one before comparing it with a complete computer. All seven cards here are ranked by what each one holds.

Models that fit in 32 GB

Every current model the roomiest 32 GB machine here holds at 32k of context, strongest first. The third column is the whole job at once: the weights plus the cache for 32k of context. The fourth says how many of the machines at this size have room for it, which is the part a size on its own cannot tell you. Speeds are on the Radeon AI PRO R9700, 32GB, the machine this list is cut from; speed follows memory bandwidth rather than memory size, so another machine at 32 GB runs the same model at its own rate, and its page gives every one of them.

For scale, the calculator starts from 80 tok/s for a hosted API and times a local machine against it. 22 of the 35 above reach it, and the slowest is 20 tok/s.

Superseded models are left out of the count, because what a machine is worth buying for is what you would run on it today; the leaderboard ranks every model this site lists, older ones included. Where the cache figure comes from is one section of the memory guide.

Qwen3 32B is one of two models this site lists at two precisions, and 32 GB holds only the lighter build. At Q4_K_M the weights plus the cache for 32k come to 28.3 GB, inside the 31 GB the roomiest machine at this size hands a model. At Q8_0 the same model needs 43.4 GB, so nothing here with 32 GB runs it. The smallest size here that does is 64 GB.

What 36 GB adds over 32 GB

The machine here with 36 GB that hands a model the most of it is the Mac Studio M5 Max, 36GB, at 27 GB, and it holds 34 of the 51. At 32 GB the most is 31 GB, on the Radeon AI PRO R9700, 32GB, holding 35. Nothing new fits. What the step buys is a longer window and room to work, not a model that was out of reach. It also gives up Nanbeige4.3 33B: the larger number on the box hands a model less here, which is why the column to read is what a model gets rather than what the box says. The strongest either way is Qwen3.8 27B.

The cheapest machine at 36 GB is the Mac Studio M5 Max, 36GB at $2,499. The cheapest at 32 GB is the Strix Halo Framework Desktop, 32GB at $1,269, so the step costs $1,230.

More memory is not more speed. The Mac Studio M5 Max, 36GB reads memory at 460 GB/s and the Radeon AI PRO R9700, 32GB at 640 GB/s, so Qwen3.8 27B runs at 19 tok/s on the 36 GB machine and 30 tok/s on the 32 GB one, both worked out from memory bandwidth.

Both sides are read off the machine at each size that hands a model the most, so the comparison is the best case against the best case for room, which is not the same as the best case for speed. The table above has every machine at 32 GB and what each of them holds.

How much context 32 GB leaves room for

The cache grows with the window you ask for, so the same machine holds fewer models the longer the context. One column here for each amount a 32 GB machine hands over. A model is counted only where its own context ceiling reaches that far.

Counted over the 51 current models, at the quantisation each is listed at and with the cache at 16 bits. The calculator has a switch for a smaller cache, which changes both what fits and how fast a long window runs.

Does a 32 GB machine pay for itself?

On the Strix Halo Framework Desktop, 32GB at $1,269, the cheapest machine at 32 GB with a published price, the model that pays it back soonest at 500k tokens a day is LFM2.5 2.6B, in 20 years. At 20M tokens a day, agents running most of the day, it is LFM2.5 2.6B in 5.9 months.

The sum is the same everywhere on this site: what the same work costs to rent, less what the electricity costs to generate it, against the price of the machine. The best buys rank the quickest pay-back at every level of use, and what it costs a month puts the machine and the API bill in the same shape.

Put your own usage in

Every figure is at 32k of context unless the row says otherwise, with the cache at 16 bits and each model at the quantisation this site lists it at. Weights are the published file sizes on each model's page, and the cache is worked out from the architecture recorded there. What a model gets is the memory the GPU can address, which each machine's page explains. Speeds say whether anybody measured them; where they were not, they are worked out from memory bandwidth. To change the context, the quantisation or the price you would pay, open the calculator. The sizes either side of this one are 24 GB and 36 GB. The memory guide has the ladder in full.