Deepseek V3.1

DeepSeek2025-08-21● 开源 · MIT🤗 deepseek-ai/DeepSeek-V3.134.6万833

DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 base with a two-phase long-context...

多服务商价格对比数据来自 OpenRouter

7 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
DeepInfra	fp4	164K	¥1.68	¥6.38	¥0.87	99.9%
S SiliconFlow	fp8	164K	¥1.81	¥6.71	-	98.2%
AtlasCloud	fp8	131K	¥2.01	¥6.71	¥0.91	99.5%
C CoreWeave	fp8	161K	¥3.69	¥11.08	¥3.69	100.0%
M Mara	—	131K	¥4.03	¥11.41	-	99.3%
Google Vertex (US West2)	—	164K	¥4.03	¥11.41	-	-
SambaNova	fp8	131K	¥4.36	¥10.07	-	96.7%

DeepInfrafp4

¥1.68

输入/M

¥6.38

输出/M

¥0.87

缓存/M

99.9%

可用率

S

SiliconFlowfp8

¥1.81

输入/M

¥6.71

输出/M

-

缓存/M

98.2%

可用率

AtlasCloudfp8

¥2.01

输入/M

¥6.71

输出/M

¥0.91

缓存/M

99.5%

可用率

C

CoreWeavefp8

¥3.69

输入/M

¥11.08

输出/M

¥3.69

缓存/M

100.0%

可用率

M

Mara

¥4.03

输入/M

¥11.41

输出/M

-

缓存/M

99.3%

可用率

Google Vertex (US West2)

¥4.03

输入/M

¥11.41

输出/M

-

缓存/M

-

可用率

SambaNovafp8

¥4.36

输入/M

¥10.07

输出/M

-

缓存/M

96.7%

可用率