How much memory does Spark-X2.6-4B need at Q4_K_M, Q8_0 and BF16?

At Q4_K_M (2.62 GB measured GGUF + 0.36 GB conversation memory at 8,192 tokens + 0.80 GB working space = 3.78 GB / 3.8 GB total estimate). At Q8_0 (4.38 GB measured GGUF + 0.36 GB conversation memory at 8,192 tokens + 0.80 GB working space = 5.54 GB / 5.5 GB total estimate). At BF16 (8.23 GB measured GGUF + 0.36 GB conversation memory at 8,192 tokens + 0.80 GB working space = 9.39 GB / 9.4 GB total estimate). XHToken's official Q4_K_M is 3.54 GB measured. With the same conversation memory and working space that file is 4.70 GB at 8,192 tokens. The bartowski Q4_K_M is the file this page uses for the standard total.

Why is Spark-X2.6-4B's conversation memory only 0.36 GB at 8,192 tokens across a 1,048,576-token max context?

Spark-X2.6-4B uses Spark2_5ForCausalLM hybrid attention: 27 sliding-window layers with a 512-token window and 9 full-attention layers. At 8,192 tokens that is 0.36 GB of conversation memory. At the full 1,048,576-token context it is 38.71 GB. The sliding layers keep only 512 tokens each, so the long context does not store a full conversation for every layer.

Where can I test open models in the browser before downloading Spark-X2.6-4B?

Spark-X2.6-4B is not one of the 10 free models. You can test those open models in the browser at https://localllmrig.com/ with no login and no local graphics memory, then read the fit on https://localllmrig.com/can-i-run/spark-x2.6-4b.

What licence does Spark-X2.6-4B use, and is commercial use allowed?

Spark-X2.6-4B is listed as Apache-2.0, and our reading of the licence is that commercial use is allowed. That is our reading of the licence, not legal advice. File sizes and the machines that hold them are on https://localllmrig.com/can-i-run/spark-x2.6-4b. You can try open models live for free at https://localllmrig.com/.

Which GPUs and Macs run Spark-X2.6-4B?

Key facts

Yes. Spark-X2.6-4B needs about 3.8 GB of graphics memory, so it runs on every graphics card setup on this page, even the RTX 3060 12GB.

About 3.8 GB at Q4_K_M, our estimate. Checked 27 Oct 2026. bartowski Q4_K_M GGUF.

The cheapest graphics card on our list that runs it (it still needs the rest of a PC): GeForce RTX 3060 12GB, $292€261, used (median of sold listings).

Fast~93 tok/sAnswers appear faster than you can read them.

Our price is for a used card. These shops mostly sell new ones, so their price can be higher.

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

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.

We never pass off an estimate as a benchmark. Commissions never change which hardware we recommend. How we calculate

The cheapest complete machine that runs it: Mac mini M4 16GB, about $615€549 (used). Comfortable~19.3 tok/s

Find hardware for Spark-X2.6-4B About this model

LIVE IN BROWSER · $0 · NO LOGIN

Test comparable open models (Recursive Flint 40B vs Gemma 4 31B) free in your browser

Open full 8-model playground ↗

Click a starter prompt below or type your own question to watch Recursive Flint stream side by side against Gemma 4 31B right on this page:

Get a 1-line alert when a new open-weight model fits 8 GB VRAM (3.8 GB at Q4 for Spark-X2.6-4B) or when used RTX 3090 / 4090 & Mac Studio prices drop >10% — weekly, zero spam.

Which graphics cards run it

Seven popular graphics cards and two PCs with two cards each, at the standard quality (Q4). Click one for the full answer: memory, speed and price.

Runs well10

The whole model fits in the graphics memory.

RTX 3060 12GBFast~93 tok/s
RTX 4060 Ti 16GBFast~74 tok/s
RTX 3090Fast~234 tok/s
RTX 4090Fast~251 tok/s
RTX 5080Fast~239 tok/s
RX 9060 XT 16GBFast~79 tok/s
RTX 5090Fast~429 tok/s
RX 7900 XTXFast~231 tok/s
Dual RTX 3090Fast~256 tok/s
Dual RTX 4090Fast~274 tok/s

Every model on every card

Which Macs and mini PCs run it

Macs and mini PCs from the picker, at the standard quality (Q4). A Mac shares its memory with the rest of the computer. Each name opens that machine's page.

Runs well12

The whole model fits in the memory the graphics can use.

Mac mini M4 16GBComfortable~19.3 tok/s
Mac mini M6 32GBFast~27 tok/s
Mac Studio M5 Max 36GBFast~73 tok/s
Mac mini M4 Pro 64GBFast~44 tok/s
Mac mini M5 Pro 64GBFast~49 tok/s
GMKtec EVO-X2 (Ryzen AI Max+ 395) 64GBFast~68 tok/s
Mac Studio M5 Ultra 96GBFast~186 tok/s
GMKtec EVO-X2 (Ryzen AI Max+ 395) 128GBFast~68 tok/s
GEEKOM A9 Mega (Ryzen AI Max+ 395) 128GBFast~68 tok/s
Beelink GTR9 Pro (Ryzen AI Max+ 395) 128GBFast~68 tok/s
Mac Studio M5 Max 128GBFast~97 tok/s
Mac Studio M5 Ultra 256GBFast~186 tok/s

Cheapest hardware for each size

From all 43 graphics cards, PCs and Macs on our list. "Runs slowly" means part of the model runs from normal memory: cheaper, but much slower. A smaller download needs less memory and answers a little less well.

Standard quality (Q4_K_M)GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Standard quality (Q4_K_M), runs slowlyNo machine on our list runs this size.
Smallest download (IQ2_M)GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Smallest download (IQ2_M), runs slowlyNo machine on our list runs this size.

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.

Prices moved: see Building a local LLM rig in the RAMageddon.

About Spark-X2.6-4B

4.11B parameters. Parameter count is an estimate. More parameters usually means a smarter model that needs more memory.

Licence:: Apache-2.0 source
Longest conversation:: about 1573 pages of text (1,048,576 tokens)
Input:: text
Released:: 19 Oct 2026 source

Sources and method

Dense text model, 4.11B parameters. Hugging Face safetensors metadata lists 4,112,079,360 parameters, all BF16. 4.11B is that count to two decimals, so our parameter count is an estimate. Architecture Spark2_5ForCausalLM (model type spark2_5): 36 layers, 27 sliding-window layers with a 512-token window and 9 full-attention layers, 4 KV heads, head dim 256, max context 1,048,576. At 8,192 tokens our estimate adds 0.36 GB of conversation memory across that 1,048,576-token max context, plus 0.80 GB of working space. Bartowski's Q4_K_M measured 2.62 GB on 27 Oct 2026 (2,621,104,768 bytes), so the total at 8,192 tokens is 3.78 GB. XHToken's official Q4_K_M measured 3.54 GB on the same day (3,543,441,312 bytes), so that file's total at 8,192 tokens is 4.70 GB. The bartowski file is the one this page uses for the standard total. Apache-2.0.

Other runtimes (not llama.cpp)

Spark2_5ForCausalLM (spark2_5) runs out of the box via GGUF in llama.cpp, Ollama and LM Studio. Stock vLLM and Transformers releases need custom model registration.

Ran Spark-X2.6-4B on your own hardware? Send us your llama-bench result.

Parameter source · Architecture source · Model page

Memory at each quality level

Each download of Spark-X2.6-4B is a different size. The total adds the file, the memory a conversation of about 12 pages of text (8,192 tokens) takes, and 0.8 GB of working space. A measured file size is labelled; the rest is our estimate. The last column is the cheapest machine that holds the total.

Swipe the table sideways to see every column, up to the cheapest machine.

Memory per quant for Spark-X2.6-4B
Quality (file)	File size	Conversation	Working space	Total	Cheapest machine that holds it
BF16	8.23 GBmeasured GGUF	0.4 GB	0.8 GB	9.4 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Q8_0	4.38 GBmeasured GGUF	0.4 GB	0.8 GB	5.5 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Q6_K_L	3.78 GBmeasured GGUF	0.4 GB	0.8 GB	4.9 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Q6_K	3.61 GBmeasured GGUF	0.4 GB	0.8 GB	4.8 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Q5_K_M	3.16 GBmeasured GGUF	0.4 GB	0.8 GB	4.3 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Q4_K_L	2.81 GBmeasured GGUF	0.4 GB	0.8 GB	4.0 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Q4_K_M	2.62 GBmeasured GGUF	0.4 GB	0.8 GB	3.8 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
IQ4_XS	2.34 GBmeasured GGUF	0.4 GB	0.8 GB	3.5 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
Q3_K_M	1.99 GBmeasured GGUF	0.4 GB	0.8 GB	3.1 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)
IQ2_M	1.57 GBmeasured GGUF	0.4 GB	0.8 GB	2.7 GB	GeForce RTX 3060 12GB, $292€261, used (median of sold listings)

Speeds and memory are our estimate from the hardware specs, not a measured test. Prices are dated and may change. How we calculate

Questions

How much memory does Spark-X2.6-4B need at Q4_K_M, Q8_0 and BF16?: At Q4_K_M (2.62 GB measured GGUF + 0.36 GB conversation memory at 8,192 tokens + 0.80 GB working space = 3.78 GB / 3.8 GB total estimate). At Q8_0 (4.38 GB measured GGUF + 0.36 GB conversation memory at 8,192 tokens + 0.80 GB working space = 5.54 GB / 5.5 GB total estimate). At BF16 (8.23 GB measured GGUF + 0.36 GB conversation memory at 8,192 tokens + 0.80 GB working space = 9.39 GB / 9.4 GB total estimate). XHToken's official Q4_K_M is 3.54 GB measured. With the same conversation memory and working space that file is 4.70 GB at 8,192 tokens. The bartowski Q4_K_M is the file this page uses for the standard total.
Why is Spark-X2.6-4B's conversation memory only 0.36 GB at 8,192 tokens across a 1,048,576-token max context?: Spark-X2.6-4B uses Spark2_5ForCausalLM hybrid attention: 27 sliding-window layers with a 512-token window and 9 full-attention layers. At 8,192 tokens that is 0.36 GB of conversation memory. At the full 1,048,576-token context it is 38.71 GB. The sliding layers keep only 512 tokens each, so the long context does not store a full conversation for every layer.
Does Spark2_5ForCausalLM run in llama.cpp, Ollama, LM Studio, vLLM or Transformers?: Spark2_5ForCausalLM (spark2_5) runs out of the box via GGUF in llama.cpp, Ollama and LM Studio. Stock vLLM and Transformers releases need custom model registration. The measured files are on https://localllmrig.com/can-i-run/spark-x2.6-4b.
Where can I test open models in the browser before downloading Spark-X2.6-4B?: Spark-X2.6-4B is not one of the 10 free models. You can test those open models in the browser at https://localllmrig.com/ with no login and no local graphics memory, then read the fit on https://localllmrig.com/can-i-run/spark-x2.6-4b.
What licence does Spark-X2.6-4B use, and is commercial use allowed?: Spark-X2.6-4B is listed as Apache-2.0, and our reading of the licence is that commercial use is allowed. That is our reading of the licence, not legal advice. File sizes and the machines that hold them are on https://localllmrig.com/can-i-run/spark-x2.6-4b. You can try open models live for free at https://localllmrig.com/.