Gemma 4 26B A4B: VRAM requirements by quant
Key facts
Gemma 4 26B A4B needs about 18.1 GB at standard quality (Q4, estimate).
Checked . unsloth UD-Q4_K_M GGUF (not Q4_K_M).
Gemma 4 26B A4B is a mid-size AI model you can download and run on your own computer. It is a mixture of experts: all of it has to fit in memory, but it uses only a small part for each word. It needs about 18 GB of memory at standard quality.
Try Gemma 4 26B A4B free in your browser, no login, no install
LIVE IN BROWSER · $0 · NO LOGIN
Test Gemma 4 26B A4B free in your browser before downloading 16.95 GB
Open full 8-model playground ↗You can test Gemma 4 26B A4B (google/gemma-4-26b-a4b-it:free, 25.2B total parameters with 3.8B active per token across 128 experts, our estimate; 262,144 max context; Apache-2.0, our reading of the licence, not legal advice) free in your browser with no login and 0 GB of local VRAM. Ask a question below or open https://localllmrig.com/ to compare Gemma 4 26B A4B side by side against dense Gemma 4 31B (20.63 GB Q4 estimate) and NVIDIA Nemotron 3.5 Lightning (19.75 GB Q4 estimate) before downloading 16.95 GB of UD-Q4_K_M weights (18.13 GB total estimate at 8,192 tokens including 0.38 GB KV cache and 0.80 GB overhead on a 24 GB GPU or 32 GB Mac).
Test Gemma 4 26B A4B vs Gemma 4 31B free in your browser →https://localllmrig.com/can-i-run/gemma-4-26b-a4b
Click a starter prompt below or type your own question to watch Gemma 4 26B A4B stream side by side against Gemma 4 31B right on this page:
Memory needed
About 18 GB
At standard quality, with room for a conversation of about 12 pages.
Get a 1-line alert when a faster quant for Gemma 4 26B A4B drops — or when GPUs that run it drop >10%
Get a 1-line alert when a new open-weight model fits 24 GB VRAM (18.1 GB at Q4 for Gemma 4 26B A4B) or when used RTX 3090 / 4090 & Mac Studio prices drop >10% — weekly, zero spam.
Dense 31B vs this mixture-of-experts model: Gemma 4 31B vs Gemma 4 26B A4B: Dense 31B vs MoE 26B-A4B on 16 GB, 24 GB & Mac. Also see Compare Claude and GPT LLMs with Local LLMs.
31B vs 26B A4B
| Gemma 4 31B | Gemma 4 26B A4B | |
|---|---|---|
| Parameters | 31.27B dense, all of them active | 25.2B total, 3.8B active |
| Measured Q4 download | 18.32 GB Q4_K_M | 16.95 GB UD-Q4_K_M (13.60 GB UD-IQ4_XS) |
| Total at 8,192 tokens | 20.63 GB | 18.13 GB (14.78 GB at UD-IQ4_XS) |
| GPU that holds it | A 24 GB GPU at Q4_K_M. A 16 GB GPU does not. | A 16 GB GPU at UD-IQ4_XS. A 24 GB GPU at UD-Q4_K_M. |
Gemma 4 31B vs Gemma 4 26B A4B. Also see Gemma 4 31B.
Machines that run Gemma 4 26B A4B
Each of these holds the whole model in memory. We do not estimate a speed for this kind of model (a mixture of experts), so these are simply the cheapest.
EU shops · Shown to visitors in Europe.US shops · Shown to visitors outside Europe.
If you buy through these links we may earn a commission. It doesn't change our ranking. How shop links work
Mac mini M6 32GB
32 GB shared memory
$1,416€1,530estimate
Machine details →Mac Studio M4 Max 36GB
36 GB shared memory · second-hand
$2,116€1,890usedunverified
Machine details →Mac mini M4 Pro 64GB
64 GB shared memory · second-hand
$2,116€1,890usedunverified
Machine details →Used PC with RTX 3090 24GB
24 GB graphics memory · second-hand
$2,413€2,346estimateused parts
Machine details →The buttons look up the graphics card, the part of this PC worth comparing.
PC with Radeon RX 7900 XTX 24GB
24 GB graphics memory · second-hand
$2,422€2,354estimateused parts
Machine details →The buttons look up the graphics card, the part of this PC worth comparing.
Where our price is for a used card or Mac, these shops mostly sell new ones, so their price can be higher.
As an Amazon Associate I earn from qualifying purchases.
Als Amazon-Partner verdiene ich an qualifizierten Verkäufen.
Als Amazon-Partner verdien ik aan in aanmerking komende aankopen.
About these prices
Each price is the amount we last wrote down for that machine, with its date on the machine page. "Estimate" is our own figure: for a complete PC we add our estimate for the other parts to the graphics card price, and some Mac prices are estimates, not a checked shop listing. "Used" is a second-hand price. "Unverified" means we have no checked shop listing for that exact amount. The shop buttons open a search for that machine at each shop, so you see today's price there. Prices excl. shipping unless marked. Euro prices incl. VAT where the seller charges it. Dollar prices are converted from our euro prices without the 21% EU VAT, before US sales tax.
Or rent one by the hour: RTX A6000 at Vast.ai holds it, for $0.40 an hour (median of live offers, 21 Oct 2026). Compare GPU cloud prices
Speeds and memory are our estimate from the hardware specs, not a measured test. Prices are dated and may change. How we calculate
Memory at each quality level
A lower quality level makes the model smaller, so it fits on a smaller machine. Standard (Q4) is what most people run.
| Quality | Memory needed |
|---|---|
| Q4 · standard (smallest that keeps quality) | 18.1 GB |
| Q5 · a bit sharper, a bit bigger | 19.1 GB |
| Q8 · near full quality, almost twice Q4 | 28.1 GB |
| FP16 · full quality, over three times Q4 | 51.6 GB |
With a conversation of 8,192 tokens · about 12 pages of text.
Sources and method
At standard quality: 17.0 GB for the model itself (a measured file), 0.4 GB for a conversation of 8,192 tokens, and 0.8 GB for the software.
Size source: huggingface.co/google/gemma-4-26B-A4B-it. 25.2B parameters in total, 3.8B used for each word. The memory counts the total.
Google's Gemma 4 mixture of experts. The model card's table lists 25.2B total and 3.8B active, with image input, under Apache 2.0. The chat's size class rounds this to 26B with 4B active. 5 of 30 layers attend to the whole context; the other 25 use a 1,024-token window. The vision projector (about 1.19 GB in BF16) is not included in these file sizes.
Licence
Commercial use: yes
Commercial use allowed. Keep the licence and copyright notice when you share the weights.
Licence: Apache-2.0 · checked 24 Oct 2026
Our reading of the licence, not legal advice.
Will Gemma 4 26B A4B run on my graphics card?
Builder interview
Roshan George (@arjie) now runs a fine-tuned Gemma 4 26B A4B, GPTQ W4A16 with group size 64, as a fast decision model on that same RTX A6000 48 GB. vLLM answers multiple-choice questions from the logprobs of one forward pass. Self-reported, 26 Oct 2026. The card is shared, so about 35 GB of the 48 GB stays available for vLLM. On Ampere (sm_86) he keeps INT4 W4A16 on Marlin kernels: weight-only FP8 measured 35 percent slower than INT4 on the same model. Read Roshan George's interview.
Common questions
- How much memory does Gemma 4 26B A4B need?
- About 18.1 GB at standard quality (Q4) with a conversation of 8,192 tokens: 17.0 GB for the model itself, 0.4 GB for the conversation and 0.8 GB for the software. The model size is a measured file; the conversation and software parts are our estimate.
- Does Gemma 4 26B A4B fit on a single 24 GB graphics card?
- Yes, by our estimate. It needs about 18.1 GB, and we count 23.4 GB of a 24 GB card as usable.
- Are the speeds measured?
- We do not show a speed for this model. It is a mixture-of-experts model, and our estimate would not be reliable for that design.
- Can I test Gemma 4 26B A4B online for free in my browser before downloading 17 GB?
- Yes. Gemma 4 26B A4B (25.2B total, 3.8B active, parameter count is an estimate; text + image in, text out) is one of the 10 free open models at https://localllmrig.com/ and in the inline test box on https://localllmrig.com/can-i-run/gemma-4-26b-a4b. You can compare it side by side against Gemma 4 31B (https://localllmrig.com/can-i-run/gemma-4-31b) or Recursive Flint (https://localllmrig.com/can-i-run/flint) for $0 with no account.
- Can I run Gemma 4 26B A4B on 24GB?
- Yes, by our estimate. At UD-Q4_K_M (16.95 GB measured GGUF + 0.38 GB conversation memory at 8,192 tokens + 0.80 GB working space = 18.13 GB / 18.1 GB total estimate), Gemma 4 26B A4B fits inside the 23.4 GB usable memory of a 24 GB card. The cheapest 24 GB card on our list is GeForce RTX 3090 24GB at €1,246 / $1,395. A 24 GB card also holds UD-Q5_K_M (21.15 GB measured GGUF, 22.3 GB total estimate).
- Can I run Gemma 4 26B A4B on a 16GB or 12GB GPU?
- Not at UD-Q4_K_M (18.1 GB total estimate) or Q4_0 (15.8 GB total estimate, 0.4 GB over the 15.4 GB usable on a 16 GB card). These measured files do fit a 16 GB GPU (15.4 GB usable; cheapest such card: Radeon RX 9060 XT 16GB at €564 / $522): UD-IQ4_XS (13.6 GB measured GGUF, 14.8 GB total estimate), UD-Q3_K_XL (12.91 GB measured GGUF, 14.1 GB total estimate), UD-IQ2_XXS (9.92 GB measured GGUF, 11.1 GB total estimate). On a 12 GB GPU (11.4 GB usable; cheapest: GeForce RTX 3060 12GB at €261 / $292), UD-IQ2_XXS (9.92 GB measured GGUF, 11.1 GB total estimate) fit. The model notes say the BF16 vision projector is about 1.19 GB and is not included in these file sizes.
- Can I run Gemma 4 26B A4B on a 32GB, 64GB, or 128GB Mac?
- Yes, by our estimate. At UD-Q4_K_M (18.1 GB total estimate) it fits on Mac mini M6 32GB (21.3 GB usable, €1,530 / $1,416, price unverified), Mac mini M4 Pro 64GB (48.0 GB usable, €1,890 / $2,116, price unverified, which also holds Q8_0 (26.86 GB measured GGUF, 28.0 GB total estimate)), and Mac Studio M4 Max 128GB and Mac Studio M5 Max 128GB (96.0 GB usable, which holds BF16 (50.51 GB measured GGUF, 51.7 GB total estimate)). It does not fit Mac mini M4 16GB at UD-Q4_K_M (18.1 GB estimate against 10.7 GB usable).
- How fast is Gemma 4 26B A4B on an RTX 5090?
- We don't give a speed for this model yet. This model only uses part of itself for each word (3.8B active of 25.2B total, a mixture of experts), and our speed estimate does not handle that yet.
- Why is Gemma 4 26B A4B's conversation memory only 0.38 GB at 8,192 tokens, and what is its license?
- 5 of its 30 layers attend to the full context (up to 262,144 tokens, about 393 pages), while the other 25 layers use a 1,024-token sliding window. That keeps the 8,192-token conversation memory to about 0.38 GB (0.4 GB rounded, our estimate). Licence: Apache-2.0.
- How does Gemma 4 26B A4B compare with Gemma 4 31B on a 16 GB or 24 GB GPU?
- Gemma 4 26B A4B is the smaller active model: a 4-bit UD-IQ4_XS file fits a 16 GB GPU, while dense Gemma 4 31B at the standard file needs a 24 GB GPU. Read https://localllmrig.com/guides/gemma-4-31b-vs-gemma-4-26b-a4b (Gemma 4 31B vs Gemma 4 26B A4B: Dense 31B vs MoE 26B-A4B on 16 GB, 24 GB & Mac) and https://localllmrig.com/guides/compare-claude-and-gpt-with-local-llms (Compare Claude and GPT LLMs with Local LLMs).
- Can I try Gemma 4 26B A4B online for free?
- Yes. Gemma 4 26B A4B is on the free chat on the home page. Pick it there, ask a question, and read the answer in your browser. No login and no install.
- Do I need a GPU to test it?
- No. The free chat runs in your browser. It does not download the model and it does not use a graphics card on your computer. A graphics card only matters if you run the model yourself.
- Do I need an account to try Gemma 4 26B A4B?
- No. The free chat needs no account and no login.