Head-to-head正面对比
DeepSeek V4 Pro 0423 vs Qwen3.7 Max
DeepSeek meets Qwen: sourced benchmark scores side by side, dimension by dimension, with prices and specs. Every conclusion below is derived from the published dataset — no opinions, no sponsorship.
DeepSeek 对 Qwen:带来源的 benchmark 成绩逐项并排,附价格与规格。下面每条结论都从公开数据集推导——没有主观意见,没有厂商赞助。
◈
The verdict, derived from data结论(全部由数据推导)
- On overall score they are nearly tied (86.3 vs 88.5) — decide on price, context and workload mix.总分几乎打平(86.3 对 88.5)——按价格、上下文和任务结构来选。
- Qwen3.7 Max is stronger in reasoning, preference.Qwen3.7 Max 在推理、偏好上更强。
● DeepSeek V4 Pro 0423 ● Qwen3.7 Max · missing dimensions count as 0 in the chart缺数据的维度在图上按 0 计
≡
Spec by spec逐项规格
| DeepSeek V4 Pro 0423 | Qwen3.7 Max | |
|---|---|---|
| Provider厂商 | DeepSeek | Qwen |
| Released发布 | 2026-04-24 | 2026-05-21 |
| Intelligence Score智能评分 | 86.3 | 88.5 |
| Benchmark coveragebenchmark 覆盖 | 95% | 88% |
| Reasoning推理 | 89.2 | 95 |
| Coding代码 | 85.9 | 85.7 |
| Knowledge知识 | 92.1 | 94.3 |
| Math数学 | 87.5 | 85.7 |
| Agent智能体 | 97.1 | 95.6 |
| Preference偏好 | 39.8 | 43.8 |
| API input $/1M输入价 $/百万 | $1.60 | $1.48 |
| API output $/1M输出价 $/百万 | $3.20 | $4.42 |
| Blended $/1M混合价 $/百万 | $2.40 | $2.95 |
| Value (score per $)性价比(分数/美元) | 36 | 30 |
| Context window上下文 | 1M | 1M |
| Max output最大输出 | 393K | 131K |
| Reasoning tiers推理档位 | thinking off / high effort / max effort | thinking on / off |
| Open weights开放权重 | Yes支持 | Yes支持 |
| CN-direct国内直连 | Yes支持 | Yes支持 |
| Free tier免费层 | No否 | No否 |
⚔
Benchmark by benchmark逐项 benchmark
| Benchmark基准 | DeepSeek V4 Pro 0423 | Qwen3.7 Max |
|---|---|---|
| Humanity's Last Exam | 48.2⚡max | 53.5 |
| GPQA Diamond | 90.1 | 92.4 |
| MMLU-Pro | 87.5 | 89.6 |
| SWE-bench Verified | 80.6 | 80.4 |
| AIME 2025 | 97⚡high | 85.7 |
| LiveCodeBench | 93.5⚡max | 91.6 |
| Terminal-Bench | 67.9⚡max | 69.7 |
| τ²-bench | 96.2 | 94.7 |
| LMArena (Chatbot Arena) | 1458 | 1474 |
DeepSeek V4 Pro 0423 dossier → 档案 → · Qwen3.7 Max dossier → 档案 → · How we score评分方法