DeepSeek V3 vs Qwen 2.5 72B
Architecture Comparison
SpecDeepSeek V3Qwen 2.5 72B
TypeMOEDENSE
Total Parameters671B72.7B
Active Parameters37B72.7B
Layers6180
Hidden Dimension7,1688,192
Attention Heads12864
KV Heads18
Context Length131,072131,072
Precision (default)BF16BF16
Total Experts256N/A
Active Experts8N/A
Memory Requirements
PrecisionDeepSeek V3Qwen 2.5 72B
BF16 Weights1342.0 GB145.4 GB
FP8 Weights671.0 GB72.7 GB
INT4 Weights335.5 GB36.4 GB
KV-Cache / Token31232 B327680 B
Activation Estimate3.00 GB2.50 GB
Minimum GPUs Needed (BF16)
H100 SXMN/A3 GPUs
L40SN/A4 GPUs
Quality Benchmarks
BenchmarkDeepSeek V3Qwen 2.5 72B
Overall8684
MMLU87.185.3
HumanEval65.056.0
GSM8K89.391.6
MT-Bench87.086.0
DeepSeek V3
MMLU
87.1
HumanEval
65.0
GSM8K
89.3
MT-Bench
87.0
Qwen 2.5 72B
MMLU
85.3
HumanEval
56.0
GSM8K
91.6
MT-Bench
86.0
Capabilities
FeatureDeepSeek V3Qwen 2.5 72B
Tool Use✓ Yes✓ Yes
Vision✗ No✗ No
Code✓ Yes✓ Yes
Math✓ Yes✓ Yes
Reasoning✗ No✗ No
Multilingual✓ Yes✓ Yes
Structured Output✓ Yes✓ Yes
API Pricing Comparison
Cheapest Output (DeepSeek V3)
$0.42/M
Input: $0.28/M
Cheapest Output (Qwen 2.5 72B)
$0.90/M
Input: $0.90/M
| Provider | DeepSeek V3 In $/M | Out $/M | Qwen 2.5 72B In $/M | Out $/M |
|---|---|---|---|---|
| deepseek | $0.28 | $0.42 | — | — |
| together | $0.50 | $2.80 | $0.90 | $0.90 |
| fireworks | — | — | $0.90 | $0.90 |
Recommendation Summary
- ‣DeepSeek V3 scores higher on overall quality (86 vs 84).
- ‣DeepSeek V3 is cheaper per output token ($0.42/M vs $0.90/M).
- ‣Qwen 2.5 72B has a smaller memory footprint (145.4 GB vs 1342.0 GB BF16), making it easier to deploy on fewer GPUs.
- ‣DeepSeek V3 uses MOE architecture while Qwen 2.5 72B uses DENSE. MoE models activate fewer parameters per token, improving inference efficiency.
- ‣DeepSeek V3 is stronger at code generation (HumanEval: 65.0 vs 56.0).
- ‣Qwen 2.5 72B is better at math reasoning (GSM8K: 91.6 vs 89.3).