·
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
| 信号 | 强度 | 权重 | 影响 |
|---|---|---|---|
| Benchmarksjust now | 83 | 30% | +25.0 |
| Pricingjust now | 100 | 15% | +14.9 |
| Recencyjust now | 96 | 15% | +14.3 |
| Capabilitiesjust now | 67 | 20% | +13.3 |
| Output Capacityjust now | 80 | 10% | +8.0 |
| Context Windowjust now | 73 | 10% | +7.3 |
LMSYS Arena Elo
1425
Percentile
87.5
Weight
30%
把当前模型放回同一服务商最近的发布节奏中查看。
社区和从业者反馈在基准测试和价格之上增加了真实世界的信号。
Share your experience with DeepSeek V3.2 and help the community make better decisions.
LMSYS Arena Elo
1425
Percentile
87.5
Weight
30%
成本估算器
每月比类别平均节省$48.61