Minimax M3

MiniMax2026-05-31● 开源 · OTHER🤗 MiniMaxAI/Minimax-M317.1万1,557

MiniMax-M3 是 MiniMax 推出的多模态基础模型,支持文本、图像、视频输入,输出文本,拥有 100 万 token 上下文窗口,适用于长周期 Agent 任务、编程等场景。

多服务商价格对比数据来自 OpenRouter

13 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
C CoreWeave	fp4	262K	¥1.54	¥6.45	¥0.34	99.2%
G GMICloud限时 -20%	fp8	1.05M	¥1.61	¥6.45	¥0.32	99.4%
DeepInfra	fp8	524K	¥1.88	¥7.39	¥0.38	99.6%
StreamLake限时 -50%	fp8	1.00M	¥2.01	¥8.06	¥0.40	97.3%
Venice	fp8	524K	¥2.01	¥8.06	¥0.40	96.1%
Together	—	524K	¥2.01	¥8.06	¥0.40	99.6%
Parasail	fp8	1.05M	¥2.01	¥8.06	¥0.40	99.3%
AtlasCloud	fp8	524K	¥2.01	¥8.06	¥0.40	99.6%
Novita	fp8	1.00M	¥2.01	¥8.06	¥0.40	99.5%
Minimax	fp8	524K	¥2.01	¥8.06	¥0.40	99.3%
M Mara	—	1.05M	¥4.03	¥16.11	-	97.1%
SambaNova	—	1.05M	¥4.03	¥16.11	¥0.40	95.4%
M ModelRun	fp4	1.05M	¥5.04	¥20.14	¥1.01	95.8%

C

CoreWeavefp4

¥1.54

输入/M

¥6.45

输出/M

¥0.34

缓存/M

99.2%

可用率

G

GMICloudfp8限时 -20%

¥1.61

输入/M

¥6.45

输出/M

¥0.32

缓存/M

99.4%

可用率

DeepInfrafp8

¥1.88

输入/M

¥7.39

输出/M

¥0.38

缓存/M

99.6%

可用率

StreamLakefp8限时 -50%

¥2.01

输入/M

¥8.06

输出/M

¥0.40

缓存/M

97.3%

可用率

Venicefp8

¥2.01

输入/M

¥8.06

输出/M

¥0.40

缓存/M

96.1%

可用率

Together

¥2.01

输入/M

¥8.06

输出/M

¥0.40

缓存/M

99.6%

可用率

Parasailfp8

¥2.01

输入/M

¥8.06

输出/M

¥0.40

缓存/M

99.3%

可用率

AtlasCloudfp8

¥2.01

输入/M

¥8.06

输出/M

¥0.40

缓存/M

99.6%

可用率

Novitafp8

¥2.01

输入/M

¥8.06

输出/M

¥0.40

缓存/M

99.5%

可用率

Minimaxfp8

¥2.01

输入/M

¥8.06

输出/M

¥0.40

缓存/M

99.3%

可用率

M

Mara

¥4.03

输入/M

¥16.11

输出/M

-

缓存/M

97.1%

可用率

SambaNova

¥4.03

输入/M

¥16.11

输出/M

¥0.40

缓存/M

95.4%

可用率

M

ModelRunfp4

¥5.04

输入/M

¥20.14

输出/M

¥1.01

缓存/M

95.8%

可用率