Head-to-head正面对比
Qwen3.7 Flash vs DeepSeek V4 Flash 0423
Qwen meets DeepSeek: 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.
Qwen 对 DeepSeek: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 (82 vs 83.4) — decide on price, context and workload mix.总分几乎打平(82 对 83.4)——按价格、上下文和任务结构来选。
- Qwen3.7 Flash is stronger in reasoning, preference.Qwen3.7 Flash 在推理、偏好上更强。
- DeepSeek V4 Flash 0423 is stronger in math.DeepSeek V4 Flash 0423 在数学上更强。
- Qwen3.7 Flash is 1.7× cheaper on blended price ($0.080 vs $0.137 per 1M tokens, in/out average).混合价(输入输出均值)上 Qwen3.7 Flash 便宜 1.7 倍:$0.080 对 $0.137/百万 token。
- Budget pick: Qwen3.7 Flash delivers 98% of the score at 59% of the blended cost.预算优先选 Qwen3.7 Flash:用 59% 的混合成本拿到 98% 的分数。
● Qwen3.7 Flash ● DeepSeek V4 Flash 0423 · 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
| Qwen3.7 Flash | DeepSeek V4 Flash 0423 | |
|---|---|---|
| Provider厂商 | Qwen | DeepSeek |
| Released发布 | 2026-07-27 | 2026-04-24 |
| Intelligence Score智能评分 | 82 | 83.4 |
| Benchmark coveragebenchmark 覆盖 | 77% | 88% |
| Reasoning推理 | 78.5 | 73.7 |
| Coding代码 | 87.3 | 88.4 |
| Knowledge知识 | 96.4 | 93.1 |
| Math数学 | 67.3 | 97 |
| Agent智能体 | — | 96.5 |
| Preference偏好 | 63 | 53 |
| API input $/1M输入价 $/百万 | $0.030 | $0.091 |
| API output $/1M输出价 $/百万 | $0.130 | $0.182 |
| Blended $/1M混合价 $/百万 | $0.080 | $0.137 |
| Value (score per $)性价比(分数/美元) | 1025 | 611 |
| Context window上下文 | 1M | 1M |
| Max output最大输出 | 66K | 944K |
| Reasoning tiers推理档位 | thinking on / off | thinking off / high effort / max effort |
| 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 | Qwen3.7 Flash | DeepSeek V4 Flash 0423 |
|---|---|---|
| Humanity's Last Exam | 35.9 | 32.1 |
| GPQA Diamond | 90.15 | 87.4 |
| MMLU-Pro | 89.311 | 86.2 |
| SWE-bench Verified | 80.4 | 79 |
| AIME 2025 | 67.3 | 97 |
| LiveCodeBench | 87.7 | 91.6⚡max |
| LMArena (Chatbot Arena) | 1475 | 1435 |
Qwen3.7 Flash dossier → 档案 → · DeepSeek V4 Flash 0423 dossier → 档案 → · How we score评分方法