微软正式推出名为MAI的七个新一代世界级人工智能模型,核心目标是在保持前沿性能的同时实现极高的价格效益比。1此次发布的模型涵盖了网络安全、编程、图像编辑、实时语音对话及转录等多个专用领域,并引入了允许企业在Azure环境中安全控制和使用模型的“MAI授权权重”机制。1通过这一机制,客户可以在自有Azure环境中部署和控制模型权重,从而保障数据隐私和知识产权安全。1
在具体的性能与成本对比中,新模型展现了显著的价格优势。1MAI Code 2在基准测试中达到了GPT-6.1 Sol性能的98%,但价格仅为后者的40%;同时,它实现了Claude Opus 5.5性能的96%,价格却只有其5%。1图像编辑模型MAI Image 2.7在Artificial Analysis图像编辑排名中位列第一,而价格仅为竞争对手的五分之一。1此外,MAI Cyber 2 Flash的输出Token价格较前代MAI Cyber 1.1 Flash降低了86%,决策路由工具Decision 1.1则比GPT-6.1 Sol快50倍且便宜79倍。1在医疗应用领域,Dragon Copilot Transcribe每月处理1000万次医患对话转录,已覆盖三分之一的美国医生。1
Microsoft has announced the release of seven new artificial intelligence models under its MAI series, emphasizing a strategy to deliver state-of-the-art capabilities at highly scalable price-performance ratios 1. The portfolio includes specialized models designed for cybersecurity, coding, image editing, and real-time voice conversation, alongside transcription services 1. A key addition to this launch is the "MAI Licensed Weights" mechanism, which allows enterprises to deploy and control model weights within their own Azure environments, ensuring data privacy and intellectual property security 1.
The specific models demonstrate significant competitive advantages in both performance and cost efficiency. MAI Cyber 2 Flash offers output token pricing that is 86% lower than its predecessor, MAI Cyber 1.1 Flash 1. In coding benchmarks, MAI Code 2 achieves 98% of the performance of GPT-6.1 Sol at only 40% of the cost, while reaching 96% of Claude Opus 5.5’s performance for just 5% of the price 1. For visual tasks, MAI Image 2.7 ranks first in the Artificial Analysis image editing leaderboard, priced at one-fifth of competitors' rates 1. Additionally, the Decision 1.1 model routing tool operates 50 times faster and is 79 times cheaper than GPT-6.1 Sol 1. The suite also features Dragon Copilot Transcribe, which currently processes 10 million doctor-patient conversation transcriptions per month, covering one-third of physicians in the United States 1.
评论
还没有评论,欢迎留下第一条。