MiniMax H3开源:中国实验室打破全模态价格

持续追踪 : Économie de l'open frontier : viabilité, subvention, pivots· 连载 13/13

模型与工具仅限订阅用户 21 min ago7加入收藏

MiniMax H3开源:中国实验室打破全模态价格
插图 : Léa Fontaine

MiniMax发布H3,全模态开源模型(文本、图像、音频、视频),价格仅为竞争对手的三分之一。AI视频市场不再是二元的封闭昂贵/开源低效。

简单来说

MiniMax 推出 H3,这是一个全模态(文本、图像、音频、视频)模型,编辑器宣布其在视频编辑领域全球排名第一,价格仅为竞争对手的三分之一。换句话说:在中国经历了6月份痛苦的股票锁定期后,这家实验室通过一个产品进行反击,该产品重新定位了其商业论点——不是前沿LLM,而是开放的多模态编排。

变化点

2026年7月31日Pandaily的公告,MiniMax的产品通信。以开源形式发布的H3模型,在同一个backbone中整合了四种模态。在视频编辑领域的领导地位声称被突出显示,但截至目前尚未有第三方基准测试。价格——比提到的竞争对手低三分之一——才是真正的市场信号:MiniMax试图通过纯粹的定位选择来降低视频价格,就像DeepSeek为文本所做的那样。

技术细节

全模态开源模型立即引发两个技术问题:权重的规模(在HuggingFace上需要关注)和实际推理。全模态模型运行一个多头路由器加上专门的heads——每秒生成视频的GPU成本取决于heads的大小,而不仅仅是backbone。显示的价格只有在自家推理时才有意义;在自托管时,实际费用将取决于硬件和集群使用情况。

结论

有两种解释。AI视频市场:7月24日的Latent Space报告将FLUX 3(Black Forest Labs)列在Seedance 2.0和Gemini Omni之前;H3在价格上低于它们,但在覆盖的模态上高于它们。市场不再是二元的(封闭昂贵/开放低端)。开放模型经济学:MiniMax押注于通过体积和硬件多样性进行补贴,就像阿里巴巴与Qwen一样。Lambert预测的封闭转向并不是唯一可行的轨迹。需要关注:创意工具(Runway、Pika、CapCut)中的实际采用情况,独立的第三方视频基准测试,以及关于权重和许可的透明度。

内容仅限会员访问

免费创建账户,即可访问我们的全部内容和每周评论。

本文由人工智能撰写,并经人工编辑审核。

我们的编辑部
Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

SSHMonitoringAI Ops
Get early access
这篇文章对您有帮助吗?

7 人赞了这篇文章

P
Priya Raman机器学习工程师
🇨🇳 机器学习工程师,应用研究
分享:
评论 (7)

登录后即可参与讨论。

Dr. Emily 31 Jul 2026 · 09:04

I wonder how this will impact the quality of content creation. Will it lead to a flood of low-quality AI-generated content or will it raise the bar for creativity?

BookWorm88 31 Jul 2026 · 08:59

I wonder how this will affect the job market for content creators. Will it lead to more opportunities or increased competition?

J.P.R. 2 31 Jul 2026 · 11:12

It might create more competition initially, but could also open up new niches for specialized content creators.

sandrine.b 31 Jul 2026 · 11:20

It might create more jobs initially, but long-term effects on the job market remain uncertain.

sandrine.b 31 Jul 2026 · 08:48

This is a game-changer! Open-source AI models are becoming more accessible and affordable. I can't wait to see what people create with this.

FilmBuffNYC 31 Jul 2026 · 08:45

I'm curious about the computational resources required to run this model. Will it be accessible to everyone or just those with high-end hardware?

HistoryBuff 31 Jul 2026 · 08:44

I'm concerned about the potential for misuse of this technology, especially in creating deepfakes or misleading content. How will this be regulated?

curio_usa 31 Jul 2026 · 08:33

This is exciting! I wonder how this will impact the creative industries, especially for small businesses and independent creators.

CriticAtHeart 31 Jul 2026 · 08:31

I wonder how this model will handle multilingual content. Will it support languages other than English effectively?

Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

Get early access
主题
浏览
信息