DeepSeek V3 vs Llama 3.1 405B
Seite an Seite. Jede Metrik. Jeder Benchmark.
| Typ | DeepSeek V3 | Llama 3.1 405B |
|---|---|---|
| Provider | ||
| Durchschn. Score | 59.0 | 38.0 |
| Eingabepreis | $0.32 | - |
| Ausgabepreis | $0.89 | - |
| Kontextfenster | 164K tokens (~82 books) | - |
| Veroeffentlicht | 2024-12-26 | 2024-07-16 |
| Open Source | Open Source | Open Source |
Benchmark-Ergebnisse
12 benchmarks · DeepSeek V3: 7, Llama 3.1 405B: 4
| Benchmark | Kategorie | DeepSeek V3 | Llama 3.1 405B |
|---|---|---|---|
| ARC AI2 | knowledge | 93.7 | 93.7 |
| BBH | reasoning | 83.3 | 77.2 |
| GPQA diamond | knowledge | 42.0 | 34.5 |
| HellaSwag | knowledge | 85.2 | 85.6 |
| MATH level 5 | math | 64.8 | 49.8 |
| MMLU | knowledge | 82.9 | 79.3 |
| OTIS Mock AIME 2024-2025 | math | 15.8 | 9.6 |
| PIQA | knowledge | 69.4 | 71.8 |
| SimpleBench | reasoning | 2.7 | 7.6 |
| TriviaQA | knowledge | 82.9 | 82.7 |
| WeirdML | coding | 36.1 | 21.4 |
| Winogrande | knowledge | 70.4 | 78.4 |