DeepSeek V3.2 vs Qwen3 30B A3B Thinking 2507
Côte à côte. Chaque métrique. Chaque benchmark.
| Type | DeepSeek V3.2 | Qwen3 30B A3B Thinking 2507 |
|---|---|---|
| Provider |  DeepSeek |  Alibaba Qwen |
| score moyen | 53.0 | 66.2 |
| Prix d'entrée | $0.26 | $0.08 |
| Prix de sortie | $0.38 | $0.40 |
| Fenêtre de contexte | 164K tokens (~82 books) | 131K tokens (~66 books) |
| Sorti le | 2025-12-01 | 2025-08-28 |
| Code source ouvert | Open Source | Open Source |
Scores de benchmark
6 benchmarks · DeepSeek V3.2: 5, Qwen3 30B A3B Thinking 2507: 0
| Benchmark | Catégorie | DeepSeek V3.2 | Qwen3 30B A3B Thinking 2507 |
|---|---|---|---|
| OpenCompass — AIME2025 | math | 93.0 | 86.8 |
| OpenCompass — GPQA-Diamond | knowledge | 84.6 | 70.1 |
| OpenCompass — HLE | knowledge | 23.2 | 11.7 |
| OpenCompass — IFEval | language | 89.7 | 89.7 |
| OpenCompass — LiveCodeBenchV6 | coding | 75.4 | 59.4 |
| OpenCompass — MMLU-Pro | knowledge | 85.8 | 79.5 |