Qwen 3 32B vs Gemma 3 27B
Architecture Comparison
SpecQwen 3 32BGemma 3 27B
TypeDENSEDENSE
Total Parameters32.8B27B
Active Parameters32.8B27B
Layers6462
Hidden Dimension5,1203,584
Attention Heads4032
KV Heads816
Context Length131,072131,072
Precision (default)BF16BF16
Memory Requirements
PrecisionQwen 3 32BGemma 3 27B
BF16 Weights65.6 GB54.0 GB
FP8 Weights32.8 GB27.0 GB
INT4 Weights16.4 GB13.5 GB
KV-Cache / Token262144 B507904 B
Activation Estimate2.00 GB1.50 GB
Minimum GPUs Needed (BF16)
H100 SXM1 GPU1 GPU
L40S2 GPUs2 GPUs
Quality Benchmarks
BenchmarkQwen 3 32BGemma 3 27B
Overall8076
MMLU82.078.0
HumanEval55.048.0
GSM8K90.085.0
MT-Bench84.082.0
Qwen 3 32B
MMLU
82.0
HumanEval
55.0
GSM8K
90.0
MT-Bench
84.0
Gemma 3 27B
MMLU
78.0
HumanEval
48.0
GSM8K
85.0
MT-Bench
82.0
Capabilities
FeatureQwen 3 32BGemma 3 27B
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 32B)
$0.80/M
Input: $0.80/M
Cheapest Output (Gemma 3 27B)
$0.20/M
Input: $0.10/M
| Provider | Qwen 3 32B In $/M | Out $/M | Gemma 3 27B In $/M | Out $/M |
|---|---|---|---|---|
| — | — | $0.10 | $0.20 | |
| together | $0.80 | $0.80 | $0.30 | $0.30 |
| fireworks | $0.90 | $0.90 | — | — |
Recommendation Summary
- ‣Qwen 3 32B scores higher on overall quality (80 vs 76).
- ‣Gemma 3 27B is cheaper per output token ($0.20/M vs $0.80/M).
- ‣Gemma 3 27B has a smaller memory footprint (54.0 GB vs 65.6 GB BF16), making it easier to deploy on fewer GPUs.
- ‣Qwen 3 32B is stronger at code generation (HumanEval: 55.0 vs 48.0).
- ‣Qwen 3 32B is better at math reasoning (GSM8K: 90.0 vs 85.0).