Llama 3.1 70B vs Llama 3.3 70B
Architecture Comparison
SpecLlama 3.1 70BLlama 3.3 70B
TypeDENSEDENSE
Total Parameters70.6B70.6B
Active Parameters70.6B70.6B
Layers8080
Hidden Dimension8,1928,192
Attention Heads6464
KV Heads88
Context Length131,072131,072
Precision (default)BF16BF16
Memory Requirements
PrecisionLlama 3.1 70BLlama 3.3 70B
BF16 Weights141.2 GB141.2 GB
FP8 Weights70.6 GB70.6 GB
INT4 Weights35.3 GB35.3 GB
KV-Cache / Token327680 B327680 B
Activation Estimate2.50 GB2.50 GB
Minimum GPUs Needed (BF16)
H100 SXM3 GPUs3 GPUs
L40S4 GPUs4 GPUs
Quality Benchmarks
BenchmarkLlama 3.1 70BLlama 3.3 70B
Overall8284
MMLU83.686.0
HumanEval58.560.0
GSM8K93.094.0
MT-Bench85.086.0
Llama 3.1 70B
MMLU
83.6
HumanEval
58.5
GSM8K
93.0
MT-Bench
85.0
Llama 3.3 70B
MMLU
86.0
HumanEval
60.0
GSM8K
94.0
MT-Bench
86.0
Capabilities
FeatureLlama 3.1 70BLlama 3.3 70B
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 (Llama 3.1 70B)
$0.79/M
Input: $0.59/M
Cheapest Output (Llama 3.3 70B)
$0.79/M
Input: $0.59/M
| Provider | Llama 3.1 70B In $/M | Out $/M | Llama 3.3 70B In $/M | Out $/M |
|---|---|---|---|---|
| groq | $0.59 | $0.79 | $0.59 | $0.79 |
| together | $0.88 | $0.88 | $0.88 | $0.88 |
| fireworks | $0.90 | $0.90 | $0.90 | $0.90 |
Recommendation Summary
- ‣Llama 3.3 70B scores higher on overall quality (84 vs 82).
- ‣Llama 3.3 70B is stronger at code generation (HumanEval: 60.0 vs 58.5).
- ‣Llama 3.3 70B is better at math reasoning (GSM8K: 94.0 vs 93.0).