Mistral Small 3.1 24B vs Gemma 3 27B
Architecture Comparison
SpecMistral Small 3.1 24BGemma 3 27B
TypeDENSEDENSE
Total Parameters24B27B
Active Parameters24B27B
Layers4062
Hidden Dimension5,1203,584
Attention Heads3232
KV Heads816
Context Length131,072131,072
Precision (default)BF16BF16
Memory Requirements
PrecisionMistral Small 3.1 24BGemma 3 27B
BF16 Weights48.0 GB54.0 GB
FP8 Weights24.0 GB27.0 GB
INT4 Weights12.0 GB13.5 GB
KV-Cache / Token163840 B507904 B
Activation Estimate1.50 GB1.50 GB
Minimum GPUs Needed (BF16)
H100 SXM1 GPU1 GPU
L40S2 GPUs2 GPUs
Quality Benchmarks
BenchmarkMistral Small 3.1 24BGemma 3 27B
Overall5076
MMLUN/A78.0
HumanEvalN/A48.0
GSM8KN/A85.0
MT-BenchN/A82.0
Mistral Small 3.1 24B
Gemma 3 27B
MMLU
78.0
HumanEval
48.0
GSM8K
85.0
MT-Bench
82.0
Capabilities
FeatureMistral Small 3.1 24BGemma 3 27B
Tool Use✓ Yes✓ Yes
Vision✓ Yes✓ Yes
Code✓ Yes✓ Yes
Math✓ Yes✓ Yes
Reasoning✗ No✗ No
Multilingual✓ Yes✓ Yes
Structured Output✓ Yes✓ Yes
API Pricing Comparison
Cheapest Output (Mistral Small 3.1 24B)
$0.30/M
Input: $0.10/M
Cheapest Output (Gemma 3 27B)
$0.20/M
Input: $0.10/M
| Provider | Mistral Small 3.1 24B In $/M | Out $/M | Gemma 3 27B In $/M | Out $/M |
|---|---|---|---|---|
| — | — | $0.10 | $0.20 | |
| mistral | $0.10 | $0.30 | — | — |
| together | — | — | $0.30 | $0.30 |
Recommendation Summary
- ‣Gemma 3 27B scores higher on overall quality (76 vs 50).
- ‣Gemma 3 27B is cheaper per output token ($0.20/M vs $0.30/M).
- ‣Mistral Small 3.1 24B has a smaller memory footprint (48.0 GB vs 54.0 GB BF16), making it easier to deploy on fewer GPUs.
Compare Other Models
Mistral Small 3.1 24B vs DeepSeek R1→Mistral Small 3.1 24B vs DeepSeek V3→Mistral Small 3.1 24B vs Llama 3.1 405B→Mistral Small 3.1 24B vs Llama 3.1 70B→Mistral Small 3.1 24B vs Llama 3.1 8B→Mistral Small 3.1 24B vs Phi-4→Gemma 3 27B vs DeepSeek R1→Gemma 3 27B vs DeepSeek V3→Gemma 3 27B vs Llama 3.1 405B→Gemma 3 27B vs Llama 3.1 70B→