Mistral Large 2 vs DeepSeek V3
Architecture Comparison
SpecMistral Large 2DeepSeek V3
TypeDENSEMOE
Total Parameters123B671B
Active Parameters123B37B
Layers8861
Hidden Dimension12,2887,168
Attention Heads96128
KV Heads81
Context Length131,072131,072
Precision (default)BF16BF16
Total ExpertsN/A256
Active ExpertsN/A8
Memory Requirements
PrecisionMistral Large 2DeepSeek V3
BF16 Weights246.0 GB1342.0 GB
FP8 Weights123.0 GB671.0 GB
INT4 Weights61.5 GB335.5 GB
KV-Cache / Token360448 B31232 B
Activation Estimate3.50 GB3.00 GB
Minimum GPUs Needed (BF16)
H100 SXM4 GPUsN/A
L40S7 GPUsN/A
Quality Benchmarks
BenchmarkMistral Large 2DeepSeek V3
Overall8286
MMLU84.087.1
HumanEval53.065.0
GSM8K91.289.3
MT-Bench84.087.0
Mistral Large 2
MMLU
84.0
HumanEval
53.0
GSM8K
91.2
MT-Bench
84.0
DeepSeek V3
MMLU
87.1
HumanEval
65.0
GSM8K
89.3
MT-Bench
87.0
Capabilities
FeatureMistral Large 2DeepSeek V3
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 (Mistral Large 2)
$2.50/M
Input: $2.50/M
Cheapest Output (DeepSeek V3)
$0.42/M
Input: $0.28/M
| Provider | Mistral Large 2 In $/M | Out $/M | DeepSeek V3 In $/M | Out $/M |
|---|---|---|---|---|
| deepseek | — | — | $0.28 | $0.42 |
| together | $2.50 | $2.50 | $0.50 | $2.80 |
| mistral | $2.00 | $6.00 | — | — |
Recommendation Summary
- ‣DeepSeek V3 scores higher on overall quality (86 vs 82).
- ‣DeepSeek V3 is cheaper per output token ($0.42/M vs $2.50/M).
- ‣Mistral Large 2 has a smaller memory footprint (246.0 GB vs 1342.0 GB BF16), making it easier to deploy on fewer GPUs.
- ‣Mistral Large 2 uses DENSE architecture while DeepSeek V3 uses MOE. MoE models activate fewer parameters per token, improving inference efficiency.
- ‣DeepSeek V3 is stronger at code generation (HumanEval: 65.0 vs 53.0).
- ‣Mistral Large 2 is better at math reasoning (GSM8K: 91.2 vs 89.3).