What can you run with 12 GB of memory?

A model does not get the 12 GB. It gets 11 GB on the one machine here sold with 12 GB, because the card keeps a margin free. That holds 14 of the 51 current models at 32k of context, and the strongest of them is Gemma 4 12B.

Run the numbers on Gemma 4 12B at 12 GB

Jump to: The machines sold with 12 GB · Models that fit in 12 GB · What 16 GB adds over 12 GB · How much context 12 GB leaves room for · Does a 12 GB machine pay for itself?

The machines sold with 12 GB

One machine here comes with 12 GB, and the fourth column is what it holds at 32k of context: the number its own page prints. No machine on sale here comes with 12 GB any more.

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. The GeForce RTX 3060, 12GB is priced as the card alone, so add the PC around it before comparing it with a complete computer. All seven cards here are ranked by what each one holds.

Models that fit in 12 GB

Every current model the one 12 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 GeForce RTX 3060, 12GB, the machine this list is cut from, which is no longer sold; speed follows memory bandwidth rather than memory size, so another machine at 12 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. 6 of the 14 above reach it, and the slowest is 29 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Superseded models are left out of the count, because what a machine is worth buying for is what you would run on it today; the leaderboard ranks every model this site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.

Llama 3.1 8B Instruct is one of two models this site lists at two precisions, and 12 GB holds only the lighter build. At Q4_K_M the weights plus the cache for 32k come to 9.2 GB, inside the 11 GB the one machine here with 12 GB hands a model. At Q8_0 the same model needs 12.8 GB, so nothing here with 12 GB runs it. The smallest size here that does is 16 GB.

What 16 GB adds over 12 GB

The machine here with 16 GB that hands a model the most of it is the GeForce RTX 4080, 16GB, at 15 GB, and it holds 16 of the 51. At 12 GB the most is 11 GB, on the GeForce RTX 3060, 12GB, holding 14. The step adds gpt-oss-20b and Ministral 3 14B. The strongest either way is Gemma 4 12B.

The cheapest machine at 16 GB is the Mac mini M6, 16GB at $899. The cheapest at 12 GB is the NVIDIA GeForce RTX 3060, 12GB at $329, card only. One of those two is a card and the other is a whole computer, so the gap between them is not the price of the step.

The step buys speed as well as room. The GeForce RTX 4080, 16GB reads memory at 716.8 GB/s and the GeForce RTX 3060, 12GB at 360 GB/s, so Gemma 4 12B runs at 74 tok/s on the 16 GB machine and 38 tok/s on the 12 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 12 GB and what each of them holds.

How much context 12 GB leaves room for

The cache grows with the window you ask for, so the same machine holds fewer models the longer the context. Every machine at this size hands a model 11 GB, so one column covers all of them. 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 12 GB machine pay for itself?

On the NVIDIA GeForce RTX 3060, 12GB at $329, card only, the cheapest machine at 12 GB with a published price, the model that pays it back soonest at 500k tokens a day is LFM2.5 2.6B, in 5.1 years. At 20M tokens a day, agents running most of the day, it is LFM2.5 2.6B in 1.5 months. The GeForce RTX 3060, 12GB is the previous generation, so that price is what it launched at.

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. This is the smallest size any machine here is sold in; the next one up is 16 GB. The memory guide has the ladder in full.