family member

Qwen 3 4B

90 tokens a second on Windows · NVIDIA 12 GB · 16 CPUs · 16 GB RAM. No hosted API sells it, so this network is the place to run it.

13 builds of Qwen 3 4B, 28 signed runs on real machines: different packagers, precisions and engines, each a different file with its own numbers.

90tok/s, fastest machine

$2.21per M tokens · no API sells it

64Kcontext proved by recall

2.0 GBmemory, smallest build

6providers measured it

Use it in your own room Ask Evo in your gramx and it runs on a machine that proved it, at the metered $2.21 per million tokens. On your own machines it costs you nothing.Measure it on your machine A gram works through the catalog by itself and signs every result. 6 providers have measured Qwen 3 4B; every run is a point on the providers board.Compare every build Put it beside every other model in one table: speed, price, context and the API price, one row each.

best on ollama · 12 builds

trained 40K · proved 32K context, by unsloth IQ3_XXS

140.7 tok/s peak, by Qwen Q4_K_M

best on vllm · 1 build

proved 64K context, by qwen

Speed against price, every machine

Up and left is faster and cheaper · the line joins machines no other beats on both · price on a log scale

Same build, different machines

IQ3_XXS by unsloth on 4 machines: the fastest decodes 286× as fast as the slowest. One dot per machine, tokens per second.

NVIDIA GeForce RTX 3080 Ti×163 tok/s
Tesla T4×149 tok/s
Intel(R) UHD Graphics 730×14.6 tok/s
Apple M4×10.2 tok/s

What it fits on

The smallest build needs about 2.0 GB; 13 of 13 kinds of machine on the network can hold it. Bars: each kind's memory (its card's, else RAM) · ticks: every build, at the size it held plus a fifth for the prompt.

NVIDIA GeForce RTX 3050 Laptop GPU4 GB on the card1 of 7
NVIDIA GeForce GTX 16504 GB on the card1 of 7
AMD Radeon(TM) Vega 8 Mobile Graphics7 GB RAM2 of 7
Windows, CPU only · 12 CPUs7 GB RAM2 of 7
apple8 GB RAM2 of 7
Apple M18 GB RAM2 of 7
NVIDIA GeForce RTX 3080 Ti12 GB on the card5 of 7
Tesla T415 GB on the cardevery build
Intel(R) UHD Graphics 73016 GB RAMevery build
Intel(R) Iris(R) Xe Graphics16 GB RAMevery build
Apple M416 GB RAMevery build
2× Tesla T429 GB on the cardevery build
Linux, CPU only · 4 CPUs31 GB RAMevery build

How far back it remembers, machine by machine

Filled: the needle came back at that prompt size · outlined: asked and lost · blank: not asked · trained for 40K. 2 walks ended because the model forgot, 1 because the machine could not hold the prompt.

machine · build	4K	8K	16K	32K	64K	128K	why it stopped
Gram ·a5be46 Tesla T4 · by qwen							64K this machine could not serve at 128K
Gram ·a5be46 Tesla T4 · IQ3_XXS by unsloth							32K recalled at every size asked
Gram ·a5be46 Tesla T4 · Q5_0 by Qwen							32K recalled at every size asked
Gram ·a5be46 Tesla T4 · Q5_K_M by Qwen							32K recalled at every size asked
Gram ·a5be46 Tesla T4 · Q8_0 by Qwen							32K recalled at every size asked
Kunal Hindocha NVIDIA GeForce RTX 3080 Ti · BF16 by unsloth							16K
Gram ·a5be46 Tesla T4 · Q4_K_M by Qwen							16K the model lost the needle at 32K
Kunal Hindocha NVIDIA GeForce RTX 3080 Ti · IQ3_XXS by unsloth							4K
nitin singh Apple M4 · Q4_0 by unsloth							4K
nitin singh Apple M4 · Q4_1 by unsloth							4K
nitin singh Apple M4 · Q5_0 by Qwen							4K
Amit Khetan Intel(R) UHD Graphics 730 · Q4_0 by unsloth							4K
Amit Khetan Intel(R) UHD Graphics 730 · IQ4_XS by unsloth							4K
nitin singh Apple M4 · IQ3_XXS by unsloth							4K

The 14 longest of 24 walks; every one is on its build's page.

Carbon per million tokens

Lowest build on each machine · 1 of 1 read off a power sensor, the rest modelled from the parts · fewer grams is better

Gram ·a5be46259 g1.98 J/tok · sensor

Carbon is location-based: signed energy × the yearly average of the grid where the machine signed its place (world grid average (place not signed); Ember Global Electricity Review 2025 (2024 generation), location-based). Facility overhead is not included. A machine on its own solar is not credited until it can prove it.

Measured by

The people whose machines signed these numbers

proved by at least one build: Tool calling · Tool loop finished · Faithful to tool results · Chained tool calls · Structured output · Sustained output · Extended thinking · Vision · Counts in images · Spatial reasoning · Writes code · Reasons about code · Synthesises sources · Follows instructions

Every build

Each its own file with its own page: the machines that ran it and their signatures.

build	engine	whose work	memory	context proved	best tok/s	machines
by qwenQwen/Qwen3-4B	vllm	qwen	—	64K	—	1
IQ3_XXS by unslothhf.co/unsloth/Qwen3-4B-GGUF:IQ3_XXS	ollama	quantizer · unsloth	13.1 GB	32K	62.9	4
Q5_0 by Qwenhf.co/Qwen/Qwen3-4B-GGUF:Q5_0	ollama	vendor · Qwen	8.8 GB	32K	48.5	3
Q5_K_M by Qwenhf.co/Qwen/Qwen3-4B-GGUF:Q5_K_M	ollama	vendor · Qwen	8.9 GB	32K	47.0	2
Q8_0 by Qwenhf.co/Qwen/Qwen3-4B-GGUF:Q8_0	ollama	vendor · Qwen	10.4 GB	32K	38.9	2
Q4_K_M by Qwenhf.co/Qwen/Qwen3-4B-GGUF:Q4_K_M	ollama	vendor · Qwen	5.7 GB	16K	140.7	2
BF16 by unslothhf.co/unsloth/Qwen3-4B-GGUF:BF16	ollama	quantizer · unsloth	12.6 GB	16K	80.8	1
Q4_0 by unslothhf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:Q4_0	ollama	unsloth	—	4K	40.9	4
Q4_1 by unslothhf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:Q4_1	ollama	unsloth	—	4K	37.0	2
IQ4_XS by unslothhf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:IQ4_XS	ollama	unsloth	—	4K	6.4	1
IQ1_S by unslothhf.co/unsloth/Qwen3-4B-GGUF:IQ1_S	ollama	quantizer · unsloth	2.0 GB	—	90.4	3
Q4_1 by unslothhf.co/unsloth/Qwen3-4B-GGUF:Q4_1	ollama	quantizer · unsloth	—	—	19.4	2
Q2_K by unslothhf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:Q2_K	ollama	unsloth	—	—	11.9	1