DeepSeek V4 Flash
DeepSeek V4 Flash 0731 是 DeepSeek 推出的稀疏混合专家(MoE)模型,总参数 2840 亿,激活参数 130 亿。这一重训练修订版适用于编程、推理与 Agent 工作流。
模态
参数量 (MoE)
284B总
激活 13B / token
上下文
1.31Mtokens
最大输出 393K tokens
能力
工具调用 (Tools)
结构化输出
并行工具调用
视觉输入
音频输入
多服务商价格对比数据来自 OpenRouter
| 服务商 | 量化精度 | 上下文 | 输入/M | 输出/M | 缓存读取/M | 近1天可用率 |
|---|---|---|---|---|---|---|
D Decart限时 -15% | fp4 | 262K | ¥0.52 | ¥1.03 | ¥0.10 | 74.6% |
| fp8 | 1.02M | ¥0.53 | ¥1.06 | ¥0.11 | 95.8% | |
| — | 1.05M | ¥0.54 | ¥1.70 | ¥0.17 | 98.3% | |
| fp8 | 1.05M | ¥0.54 | ¥1.22 | ¥0.11 | 98.1% | |
O OpenInference | fp4 | 262K | ¥0.54 | ¥1.22 | ¥0.19 | 95.9% |
S Sail Research | fp4 | 262K | ¥0.61 | ¥1.22 | ¥0.14 | 98.1% |
| fp4 | 1.05M | ¥0.71 | ¥1.42 | ¥0.14 | 97.4% | |
G GMICloud限时 -20% | fp8 | 1.05M | ¥0.76 | ¥1.51 | ¥0.15 | 98.4% |
| fp8 | 1.05M | ¥0.88 | ¥1.76 | ¥0.19 | 98.8% | |
C CoreWeave | fp8 | 262K | ¥0.88 | ¥1.89 | ¥0.47 | 99.8% |
I Inceptron | fp4 | 1.05M | ¥0.88 | ¥1.89 | ¥0.20 | 98.2% |
| bf16 | 1.05M | ¥0.94 | ¥1.88 | ¥0.47 | 95.6% | |
| — | 1.05M | ¥0.95 | ¥1.89 | ¥0.19 | 94.9% | |
A AkashML | fp8 | 131K | ¥0.95 | ¥1.89 | ¥0.14 | 95.5% |
| fp8 | 1.05M | ¥0.95 | ¥1.89 | ¥0.19 | 98.9% | |
| — | 1.05M | ¥0.95 | ¥1.89 | ¥0.20 | 95.6% | |
| fp8 | 1.05M | ¥0.95 | ¥1.89 | ¥0.47 | 97.5% | |
| fp4 | 1.05M | ¥0.95 | ¥1.89 | ¥0.19 | 99.3% | |
S SiliconFlow | fp8 | 1.05M | ¥0.95 | ¥1.89 | ¥0.19 | 97.9% |
A Ambient | fp4 | 1.05M | ¥0.95 | ¥1.89 | ¥0.19 | 94.1% |
| fp8 | 1.05M | ¥0.95 | ¥1.89 | ¥0.19 | 98.4% | |
M Mancer 2 | fp8 | 1.05M | ¥0.98 | ¥3.04 | - | 95.8% |
I Io Net | fp8 | 262K | ¥1.01 | ¥2.16 | ¥0.52 | 98.4% |
| — | 1.00M | ¥1.18 | ¥2.36 | ¥0.24 | 94.1% | |
P Phala | — | 1.05M | ¥1.35 | ¥2.70 | ¥0.47 | 97.1% |
| — | 1.05M | ¥1.49 | ¥4.46 | ¥0.05 | 96.5% | |
W Wafer (Fast) | — | 1.05M | ¥1.89 | ¥3.78 | ¥0.47 | 98.7% |
| — | 1.31M | ¥2.97 | ¥8.92 | ¥0.09 | 99.9% |
D
Decartfp4限时 -15%¥0.52
输入/M
¥1.03
输出/M
¥0.10
缓存/M
74.6%
可用率
¥0.53
输入/M
¥1.06
输出/M
¥0.11
缓存/M
95.8%
可用率
¥0.54
输入/M
¥1.70
输出/M
¥0.17
缓存/M
98.3%
可用率
¥0.54
输入/M
¥1.22
输出/M
¥0.11
缓存/M
98.1%
可用率
O
OpenInferencefp4¥0.54
输入/M
¥1.22
输出/M
¥0.19
缓存/M
95.9%
可用率
S
Sail Researchfp4¥0.61
输入/M
¥1.22
输出/M
¥0.14
缓存/M
98.1%
可用率
¥0.71
输入/M
¥1.42
输出/M
¥0.14
缓存/M
97.4%
可用率
G
GMICloudfp8限时 -20%¥0.76
输入/M
¥1.51
输出/M
¥0.15
缓存/M
98.4%
可用率
¥0.88
输入/M
¥1.76
输出/M
¥0.19
缓存/M
98.8%
可用率
C
CoreWeavefp8¥0.88
输入/M
¥1.89
输出/M
¥0.47
缓存/M
99.8%
可用率
I
Inceptronfp4¥0.88
输入/M
¥1.89
输出/M
¥0.20
缓存/M
98.2%
可用率
¥0.94
输入/M
¥1.88
输出/M
¥0.47
缓存/M
95.6%
可用率
¥0.95
输入/M
¥1.89
输出/M
¥0.19
缓存/M
94.9%
可用率
A
AkashMLfp8¥0.95
输入/M
¥1.89
输出/M
¥0.14
缓存/M
95.5%
可用率
¥0.95
输入/M
¥1.89
输出/M
¥0.19
缓存/M
98.9%
可用率
¥0.95
输入/M
¥1.89
输出/M
¥0.20
缓存/M
95.6%
可用率
¥0.95
输入/M
¥1.89
输出/M
¥0.47
缓存/M
97.5%
可用率
¥0.95
输入/M
¥1.89
输出/M
¥0.19
缓存/M
99.3%
可用率
S
SiliconFlowfp8¥0.95
输入/M
¥1.89
输出/M
¥0.19
缓存/M
97.9%
可用率
A
Ambientfp4¥0.95
输入/M
¥1.89
输出/M
¥0.19
缓存/M
94.1%
可用率
¥0.95
输入/M
¥1.89
输出/M
¥0.19
缓存/M
98.4%
可用率
M
Mancer 2fp8¥0.98
输入/M
¥3.04
输出/M
-
缓存/M
95.8%
可用率
I
Io Netfp8¥1.01
输入/M
¥2.16
输出/M
¥0.52
缓存/M
98.4%
可用率
¥1.18
输入/M
¥2.36
输出/M
¥0.24
缓存/M
94.1%
可用率
P
Phala¥1.35
输入/M
¥2.70
输出/M
¥0.47
缓存/M
97.1%
可用率
¥1.49
输入/M
¥4.46
输出/M
¥0.05
缓存/M
96.5%
可用率
W
Wafer (Fast)¥1.89
输入/M
¥3.78
输出/M
¥0.47
缓存/M
98.7%
可用率
¥2.97
输入/M
¥8.92
输出/M
¥0.09
缓存/M
99.9%
可用率
📊 Artificial Analysis 综合指数
51.8
智能指数
69.1
编程指数
48.4
Agent 指数