DeepSeek R1 vs Qwen 3 235B
Architecture Comparison
SpecDeepSeek R1Qwen 3 235B
TypeMOEMOE
Total Parameters671B235B
Active Parameters37B22B
Layers6194
Hidden Dimension7,1685,120
Attention Heads12864
KV Heads14
Context Length131,072131,072
Precision (default)BF16BF16
Total Experts256128
Active Experts88
Memory Requirements
PrecisionDeepSeek R1Qwen 3 235B
BF16 Weights1342.0 GB470.0 GB
FP8 Weights671.0 GB235.0 GB
INT4 Weights335.5 GB117.5 GB
KV-Cache / Token31232 B192512 B
Activation Estimate3.00 GB3.00 GB
Minimum GPUs Needed (BF16)
H100 SXMN/A7 GPUs
L40SN/AN/A
Quality Benchmarks
BenchmarkDeepSeek R1Qwen 3 235B
Overall9288
MMLU90.888.0
HumanEval71.762.0
GSM8K97.394.0
MT-Bench89.088.0
DeepSeek R1
MMLU
90.8
HumanEval
71.7
GSM8K
97.3
MT-Bench
89.0
Qwen 3 235B
MMLU
88.0
HumanEval
62.0
GSM8K
94.0
MT-Bench
88.0
Capabilities
FeatureDeepSeek R1Qwen 3 235B
Tool Use✓ Yes✓ Yes
Vision✗ No✗ No
Code✓ Yes✓ Yes
Math✓ Yes✓ Yes
Reasoning✓ Yes✓ Yes
Multilingual✓ Yes✓ Yes
Structured Output✓ Yes✓ Yes
API Pricing Comparison
Cheapest Output (DeepSeek R1)
$2.19/M
Input: $0.55/M
Cheapest Output (Qwen 3 235B)
$3.00/M
Input: $1.50/M
| Provider | DeepSeek R1 In $/M | Out $/M | Qwen 3 235B In $/M | Out $/M |
|---|---|---|---|---|
| deepseek | $0.55 | $2.19 | — | — |
| together | $3.00 | $7.00 | $1.50 | $3.00 |
| fireworks | — | — | $1.80 | $3.50 |
Recommendation Summary
- ‣DeepSeek R1 scores higher on overall quality (92 vs 88).
- ‣DeepSeek R1 is cheaper per output token ($2.19/M vs $3.00/M).
- ‣Qwen 3 235B has a smaller memory footprint (470.0 GB vs 1342.0 GB BF16), making it easier to deploy on fewer GPUs.
- ‣DeepSeek R1 is stronger at code generation (HumanEval: 71.7 vs 62.0).
- ‣DeepSeek R1 is better at math reasoning (GSM8K: 97.3 vs 94.0).