DeepSeek V3.2 Speciale vs Llama 3.1 Nemotron Ultra 253B v1
DeepSeek vs NVIDIA — benchmarks, pricing, and capabilities side by side.
- •DeepSeek V3.2 Speciale has the higher intelligence index (89.9 vs 78.3)
- •DeepSeek V3.2 Speciale is cheaper ($0.29 vs $0.60 per 1M input)
| DeepSeek V3.2 Speciale | Llama 3.1 Nemotron Ultra 253B v1 | |
|---|---|---|
| Intelligence index | 89.9 | 78.3 |
| Developer | DeepSeek | NVIDIA |
| Type | LLM | LLM |
| Access | Open weights | Open weights |
| Context window | 163,840 tokens | — |
| Input price | $0.29 / 1M | $0.60 / 1M |
| Output price | $0.43 / 1M | $1.80 / 1M |
| Speed | — | 42 tok/s |
| Released | December 1, 2025 | April 7, 2025 |
| Parameters | — | 253000000000 |
| Input modalities | Text | — |
| Output modalities | Text | — |
Shared benchmarks
DeepSeek V3.2 Speciale
Llama 3.1 Nemotron Ultra 253B v1
AIME 2025
96.7
72.5
GPQA Diamond
87.1
76
Humanity’s Last Exam
26.1
8.1
LiveCodeBench
89.6
66.3
MMLU-Pro
86.3
82.5