Arena AI 中文镜像
CNDeepSeek

DeepSeek V4 Flash

DeepSeek2026-07-31● 开源 · MIT🤗 deepseek-ai/DeepSeek-V4-Flash-0731449.5万4,012

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
1.05Mtokens
最大输出 944K tokens
排名与分数走势
Deepseek V4 FlashDeepseek V4 Flash High (Preview)
排名
#61#77#92#1086/197/248/2810/3
Elo 评分
14011425144814726/197/248/2810/3
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
28 个服务商1 USD ≈ 6.7144 CNY · 更新于 2026-10-03
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
RelaceRelacefp41.05M¥0.10¥8.59¥0.0999.7%
O
OpenInference
fp41.05M¥0.11¥8.59¥0.1099.9%
S
Sail Research
fp41.05M¥0.13¥2.01¥0.0998.4%
S
Sail Research
fp41.05M¥0.13¥2.82¥0.0896.4%
R
Reka
—262K¥0.14¥3.55¥0.0499.8%
StreamLakeStreamLake限时 -90%fp81.02M¥0.30¥0.89¥0.0199.7%
I
Inceptron
fp41.05M¥0.34¥4.36¥0.1899.8%
DeepInfraDeepInfrafp81.05M¥0.40¥1.21¥0.1099.8%
DigitalOceanDigitalOcean—1.05M¥0.80¥1.60¥0.1699.7%
W
Wafer (Fast)
—1.05M¥0.81¥6.71¥0.34100.0%
BaseTenBaseTenfp81.05M¥0.87¥1.75¥0.19100.0%
BaseTenBaseTenfp81.05M¥0.87¥1.75¥0.19100.0%
C
CoreWeave
fp8262K¥0.87¥1.88¥0.47100.0%
CohereCohere—1.05M¥0.94¥1.88¥0.4799.5%
TogetherTogether—1.05M¥0.94¥1.88¥0.2099.6%
ParasailParasailfp81.05M¥0.94¥1.88¥0.3499.7%
MorphMorphbf161.05M¥0.95¥2.69¥0.2498.1%
VeniceVenice—1.00M¥1.18¥2.35¥0.2499.5%
M
Mancer 2
fp81.05M¥1.34¥4.03-99.2%
S
SiliconFlow
fp81.05M¥1.48¥4.43¥0.1999.4%
G
GMICloud限时 -35%
fp81.05M¥1.92¥5.76¥0.06100.0%
N
NextBit
fp81.05M¥2.36¥7.09¥0.08100.0%
AlibabaAlibaba 时段定价—1.00M¥2.36¥7.09¥0.2499.0%
NovitaNovita限时 -7%fp81.05M¥2.75¥8.24¥0.17100.0%
P
Phala
—1.05M¥2.95¥8.86¥0.1999.8%
BaiduBaidufp81.05M¥2.95¥8.86¥0.0999.9%
AtlasCloudAtlasCloudfp41.05M¥2.95¥8.86¥0.1999.9%
CloudflareCloudflare—1.05M¥2.95¥8.86¥0.09100.0%
RelaceRelacefp4
¥0.10
输入/M
¥8.59
输出/M
¥0.09
缓存/M
99.7%
可用率
O
OpenInferencefp4
¥0.11
输入/M
¥8.59
输出/M
¥0.10
缓存/M
99.9%
可用率
S
Sail Researchfp4
¥0.13
输入/M
¥2.01
输出/M
¥0.09
缓存/M
98.4%
可用率
S
Sail Researchfp4
¥0.13
输入/M
¥2.82
输出/M
¥0.08
缓存/M
96.4%
可用率
R
Reka
¥0.14
输入/M
¥3.55
输出/M
¥0.04
缓存/M
99.8%
可用率
StreamLakeStreamLakefp8限时 -90%
¥0.30
输入/M
¥0.89
输出/M
¥0.01
缓存/M
99.7%
可用率
I
Inceptronfp4
¥0.34
输入/M
¥4.36
输出/M
¥0.18
缓存/M
99.8%
可用率
DeepInfraDeepInfrafp8
¥0.40
输入/M
¥1.21
输出/M
¥0.10
缓存/M
99.8%
可用率
DigitalOceanDigitalOcean
¥0.80
输入/M
¥1.60
输出/M
¥0.16
缓存/M
99.7%
可用率
W
Wafer (Fast)
¥0.81
输入/M
¥6.71
输出/M
¥0.34
缓存/M
100.0%
可用率
BaseTenBaseTenfp8
¥0.87
输入/M
¥1.75
输出/M
¥0.19
缓存/M
100.0%
可用率
BaseTenBaseTenfp8
¥0.87
输入/M
¥1.75
输出/M
¥0.19
缓存/M
100.0%
可用率
C
CoreWeavefp8
¥0.87
输入/M
¥1.88
输出/M
¥0.47
缓存/M
100.0%
可用率
CohereCohere
¥0.94
输入/M
¥1.88
输出/M
¥0.47
缓存/M
99.5%
可用率
TogetherTogether
¥0.94
输入/M
¥1.88
输出/M
¥0.20
缓存/M
99.6%
可用率
ParasailParasailfp8
¥0.94
输入/M
¥1.88
输出/M
¥0.34
缓存/M
99.7%
可用率
MorphMorphbf16
¥0.95
输入/M
¥2.69
输出/M
¥0.24
缓存/M
98.1%
可用率
VeniceVenice
¥1.18
输入/M
¥2.35
输出/M
¥0.24
缓存/M
99.5%
可用率
M
Mancer 2fp8
¥1.34
输入/M
¥4.03
输出/M
-
缓存/M
99.2%
可用率
S
SiliconFlowfp8
¥1.48
输入/M
¥4.43
输出/M
¥0.19
缓存/M
99.4%
可用率
G
GMICloudfp8限时 -35%
¥1.92
输入/M
¥5.76
输出/M
¥0.06
缓存/M
100.0%
可用率
N
NextBitfp8
¥2.36
输入/M
¥7.09
输出/M
¥0.08
缓存/M
100.0%
可用率
AlibabaAlibaba 时段定价
¥2.36
输入/M
¥7.09
输出/M
¥0.24
缓存/M
99.0%
可用率
NovitaNovitafp8限时 -7%
¥2.75
输入/M
¥8.24
输出/M
¥0.17
缓存/M
100.0%
可用率
P
Phala
¥2.95
输入/M
¥8.86
输出/M
¥0.19
缓存/M
99.8%
可用率
BaiduBaidufp8
¥2.95
输入/M
¥8.86
输出/M
¥0.09
缓存/M
99.9%
可用率
AtlasCloudAtlasCloudfp4
¥2.95
输入/M
¥8.86
输出/M
¥0.19
缓存/M
99.9%
可用率
CloudflareCloudflare
¥2.95
输入/M
¥8.86
输出/M
¥0.09
缓存/M
100.0%
可用率
📊 Artificial Analysis 综合指数
34.3
智能指数
69.1
编程指数
41
Agent 指数