Spark X2.6 4B Q4_K_M
on Mac, M4 Pro, 48 GB

Fits with about 41 GB to spare. Finished 3 of 15 runs inside a 600 token limit.

Download this entryAs a CSV file
How we testHow a figure gets in
Run it yourselfThe script, step by step
Cite this entry
Follow Onticpost

Run history

Every run

Date	Speed	Memory
30 October 2026	64.1 t/s	7.5 GB

Change over time

How each figure moved, from the earliest run in each period to the latest.

The table fills once a second run is recorded.

Did this match your machine?

Readers who ran the script tell us whether they saw the same.

No answers yet. Be the first.

Against the maker's claim

The maker publishes no speed or memory figure for this build, so there is nothing to set our figures against.

How we measured

The set up for this entry has not been written down yet.

How every figure in the index is measured

Run it yourself

Send us your result in the same format and it joins the index as a community entry, with your sample size.

curl -fsSL https://onticpost.com/run.sh | sh -s -- \ --model MODEL --runs 6

The script needs Ollama or LM Studio running on your machine, with the model loaded. It changes nothing and sends nothing; it prints the line to send us.

About Spark X2.6 4B Q4_K_M

What is Spark X2.6 4B?

Spark X2.6 4B is a compact language model from XHToken for on device agents, with three thinking levels (none, low and high), released under the Apache 2.0 licence. This entry measures the Q4_K_M build, a 3.54 GB file, in LM Studio on a Mac with an M4 Pro chip and 48 GB of memory.

What was measured

Fifteen runs across five tasks in LM Studio, llama.cpp Metal runtime 2.47.0, context 65,536 tokens, 600 tokens allowed per run, on macOS 27.0.1. Writing speed is the middle run. Memory is the larger of the readings taken before and after the runs. 3 of the 15 runs finished inside the allowance. In 12 the whole allowance went on thinking and no answer was written.

Other figures

Measure	Result
Slowest and fastest run	56.15 and 67.33 tokens a second
Wait for the first token, short prompt	0.19 seconds
Wait for the first token, 3,862 token prompt	7.0 seconds
Runs that finished inside 600 tokens	3 of 15
Runs with no answer inside 600 tokens	12 of 15
Model file on disk	3.54 GB

Measured with the same script as every other entry on this machine.

Sources

XHToken's Spark X2.6 4B model card

Cite this page

Hill. "Spark X2.6 4B Q4_K_M". Onticpost, published 3 November 2026, updated 31 October 2026. https://onticpost.com/index/spark-x2-6-4b-m4-pro-48gb/

Updated 31 October 2026. This address stays the same when the piece is updated.

Similar entries: the same model

No other machine has run this model in the index yet. Run the script above and yours can be the first.

Similar entries: this machine

Entry	Speed	Fit
Nemotron 3.5 Bolt E4B Q4_K_M	52.1
Nanbeige4.3 33B Q4_K_M	9.7
K2 Horizon 0.9B Q4_K_M	224.7
MiMo V2.6 Distill Qwen 9B Q8_0	24.3
Flint Q4_K_M	71.5
GLM 4.7 Flash Q4_K_M	56.2