DeepSeek V3.2 vs Qwen3 235B A22B Instruct 2507
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | DeepSeek V3.2 | Qwen3 235B A22B Instruct 2507 |
|---|---|---|
| Provider |  DeepSeek |  Alibaba Qwen |
| pontuação média | 53.0 | 48.5 |
| Preço de entrada | $0.26 | $0.07 |
| Preço de saída | $0.38 | $0.10 |
| Janela de contexto | 164K tokens (~82 books) | 262K tokens (~131 books) |
| Lançado em | 2025-12-01 | 2025-07-21 |
| Código aberto | Open Source | Open Source |
Pontuações de benchmark
18 benchmarks · DeepSeek V3.2: 15, Qwen3 235B A22B Instruct 2507: 3
| Benchmark | Categoria | DeepSeek V3.2 | Qwen3 235B A22B Instruct 2507 |
|---|---|---|---|
| Aider polyglot | coding | 74.2 | 59.6 |
| ARC-AGI | reasoning | 57.0 | 11.0 |
| ARC-AGI-2 | reasoning | 4.0 | 1.3 |
| Chatbot Arena Elo — Overall | arena | 1424.4 | 1422.6 |
| LiveBench — Agentic Coding | coding | 46.7 | 13.3 |
| LiveBench — Coding | coding | 75.7 | 69.6 |
| LiveBench — Data Analysis | reasoning | 45.0 | 44.7 |
| LiveBench — If | language | 23.1 | 21.7 |
| LiveBench — Language | language | 64.2 | 66.1 |
| LiveBench — Mathematics | math | 64.0 | 68.0 |
| LiveBench — Overall | knowledge | 51.8 | 48.8 |
| LiveBench — Reasoning | reasoning | 44.3 | 58.4 |
| OpenCompass — AIME2025 | math | 93.0 | 69.5 |
| OpenCompass — GPQA-Diamond | knowledge | 84.6 | 75.5 |
| OpenCompass — HLE | knowledge | 23.2 | 12.3 |
| OpenCompass — IFEval | language | 89.7 | 88.3 |
| OpenCompass — LiveCodeBenchV6 | coding | 75.4 | 43.0 |
| OpenCompass — MMLU-Pro | knowledge | 85.8 | 79.2 |