Qwen3-4B

Qwen family · 4.0B params

2recorded runs

80.2best tok/s

0verified

Evidence: 0 sourced runs · 2 awaiting review · 1 independent source · first recorded Aug 16, 2026 · last recorded Aug 16, 2026. What these labels mean.

What the Qwen3-4B numbers mean

The fastest result on our board is 80.2 output tok/s on an Intel Arc Pro B70 at Q4_K_M with a 2048-token context. That is about 12.5 ms per generated token — 500 tokens takes roughly 6.2 seconds. Prompt processing on the same record runs at 2395.8 tok/s, so a 2,000-token prompt is read in about 0.8 seconds. Across the 2 records we hold for this model the range is 55.3–80.2 tok/s on 2 GPUs, because quantisation, context length and the engine build all move the number.

How much VRAM does Qwen3-4B need?

Qwen3-4B has 4.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	3.4 GB	8 GB
Q8_0 (8-bit)	5.7 GB	8 GB
FP16 (16-bit)	10.0 GB	12 GB

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Fastest GPUs for this model (best tok/s)

Intel Arc Pro B7080.2
Intel Arc Pro B6055.3

Models with a similar footprint

Model	Params	Q4_K_M weights
gemma-3n-E4B-it-text	4B	2.2 GB
NVIDIA-Nemotron-3-Nano-4B	4B	2.2 GB
Phi-4-mini-reasoning	4B	2.2 GB
Qwen3-4B-Thinking-2507	4B	2.2 GB
Qwen3.5-4B	4B	2.2 GB
Qwen3.5-4B-MTP	4B	2.2 GB

Ranked by distance from Qwen3-4B's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
Intel Arc Pro B70 · 32 GB	Qwen3-4B	Q4_K_M	llama.cpp —	2048	80.2 …	2395.8	—	src
Intel Arc Pro B60 · 24 GB	Qwen3-4B	Q4_K_M	llama.cpp —	2048	55.3 …	1698.3	—	src