Qwen 3 4B vs Gemma 3 4B
Architecture Comparison
SpecQwen 3 4BGemma 3 4B
TypeDENSEDENSE
Total Parameters4B4.3B
Active Parameters4B4.3B
Layers3634
Hidden Dimension2,5602,560
Attention Heads3232
KV Heads88
Context Length131,072131,072
Precision (default)BF16BF16
Memory Requirements
PrecisionQwen 3 4BGemma 3 4B
BF16 Weights8.0 GB8.6 GB
FP8 Weights4.0 GB4.3 GB
INT4 Weights2.0 GB2.1 GB
KV-Cache / Token147456 B139264 B
Activation Estimate0.50 GB0.50 GB
Minimum GPUs Needed (BF16)
H100 SXM1 GPU1 GPU
L40S1 GPU1 GPU
Quality Benchmarks
BenchmarkQwen 3 4BGemma 3 4B
Overall5754
MMLU64.060.0
HumanEval35.032.0
GSM8K65.058.0
MT-Bench73.072.0
Qwen 3 4B
MMLU
64.0
HumanEval
35.0
GSM8K
65.0
MT-Bench
73.0
Gemma 3 4B
MMLU
60.0
HumanEval
32.0
GSM8K
58.0
MT-Bench
72.0
Capabilities
FeatureQwen 3 4BGemma 3 4B
Tool Use✓ Yes✓ Yes
Vision✗ No✓ Yes
Code✓ Yes✓ Yes
Math✓ Yes✓ Yes
Reasoning✓ Yes✗ No
Multilingual✓ Yes✓ Yes
Structured Output✓ Yes✓ Yes
API Pricing Comparison
Cheapest Output (Qwen 3 4B)
$0.10/M
Input: $0.10/M
Cheapest Output (Gemma 3 4B)
$0.10/M
Input: $0.05/M
| Provider | Qwen 3 4B In $/M | Out $/M | Gemma 3 4B In $/M | Out $/M |
|---|---|---|---|---|
| together | $0.10 | $0.10 | — | — |
| — | — | $0.05 | $0.10 |
Recommendation Summary
- ‣Qwen 3 4B scores higher on overall quality (57 vs 54).
- ‣Qwen 3 4B has a smaller memory footprint (8.0 GB vs 8.6 GB BF16), making it easier to deploy on fewer GPUs.
- ‣Qwen 3 4B is stronger at code generation (HumanEval: 35.0 vs 32.0).
- ‣Qwen 3 4B is better at math reasoning (GSM8K: 65.0 vs 58.0).