Gemma 3 27B vs Qwen 3 32B
Architecture Comparison
SpecGemma 3 27BQwen 3 32B
TypeDENSEDENSE
Total Parameters27B32.8B
Active Parameters27B32.8B
Layers6264
Hidden Dimension3,5845,120
Attention Heads3240
KV Heads168
Context Length131,072131,072
Precision (default)BF16BF16
Memory Requirements
PrecisionGemma 3 27BQwen 3 32B
BF16 Weights54.0 GB65.6 GB
FP8 Weights27.0 GB32.8 GB
INT4 Weights13.5 GB16.4 GB
KV-Cache / Token507904 B262144 B
Activation Estimate1.50 GB2.00 GB
Minimum GPUs Needed (BF16)
H100 SXM1 GPU1 GPU
L40S2 GPUs2 GPUs
Quality Benchmarks
BenchmarkGemma 3 27BQwen 3 32B
Overall7680
MMLU78.082.0
HumanEval48.055.0
GSM8K85.090.0
MT-Bench82.084.0
Gemma 3 27B
MMLU
78.0
HumanEval
48.0
GSM8K
85.0
MT-Bench
82.0
Qwen 3 32B
MMLU
82.0
HumanEval
55.0
GSM8K
90.0
MT-Bench
84.0
Capabilities
FeatureGemma 3 27BQwen 3 32B
Tool Use✓ Yes✓ Yes
Vision✓ Yes✗ No
Code✓ Yes✓ Yes
Math✓ Yes✓ Yes
Reasoning✗ No✓ Yes
Multilingual✓ Yes✓ Yes
Structured Output✓ Yes✓ Yes
API Pricing Comparison
Cheapest Output (Gemma 3 27B)
$0.20/M
Input: $0.10/M
Cheapest Output (Qwen 3 32B)
$0.80/M
Input: $0.80/M
| Provider | Gemma 3 27B In $/M | Out $/M | Qwen 3 32B In $/M | Out $/M |
|---|---|---|---|---|
| $0.10 | $0.20 | — | — | |
| together | $0.30 | $0.30 | $0.80 | $0.80 |
| 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).