微软在 Ignite 2026 发布 7 款 MAI 新模型
Ignite 2026: Microsoft AI Announces 7 New Models
微软 AI 在 Ignite 2026 大会上发布 7 款新模型,包括 MAI Cyber 2 Flash、MAI Code 2、MAI Code 2 Flash、MAI Image 2.7、MAI Realtime 1、Dragon Copilot Transcribe 和 Decision 1.1。微软称 MAI Cyber 2 Flash 质量与 GPT-6 Astra 相当而每任务成本不到 2%,MAI Code 2 达到 GPT-6.1 Sol 编码质量的 98%、价格约为 40%。这些模型已在第二代 Maia 200 加速器上投入生产,微软还推出 MAI Licensed Weights,允许企业在 Azure 上部署并定制模型。
正文
At the Ignite 2026 conference today, Microsoft AI announced 7 new “world class” and customizable AI models.
“Our mission at MAI is to build you the very best models in the world, and to keep every developer ad organization right at the absolute frontier of capability,” Microsoft AI CEO Mustafa Suleyman said during the Ignite keynote today. “Our commitment is to give you the very best price performance out there so you can really scale your agents … We don’t just hill-climb on frontier quality alone. We co-optimize quality and price together.”
The new models are:
MAI Cyber 2 Flash, which Microsoft Ai says matches GPT-6 Astra in quality at less than 2 percent of the cost per task.
MAI Code 2, which provides 98 percent of the coding quality of GPT-6.1 Sol at about 40 percent of the price. Against Opus 5.5, this model reaches 96 percent of the performance for just 5 percent of the price.
MAI Code 2 Flash, which matches of the performance of GPT-6 Luna at 65 percent of the price, and Haiku 5.5 at 20 percent of the price.
MAI Image 2.7, which is now the top-rated model on Artificial Analysis fr image editing, outperforming GPT Image 2.6 Sunburst (max) at just one fifth of the price.
MAI Realtime 1, which outperforms GPT-Realtime 2.1 and starts at just $0.01 per audio minute.
Dragon Copilot Transcribe, which doesn’t get a dedicated blog post for some reason. It’s built on MAI Transcribe 2, which MAI says is the world’s fastest and most accurate multilingual transcription model, and it already powers 10 million patient encounters a month with multilingual error rates halved.
Decision 1.1 is the latest decision model from Microsoft, while being 50x faster and 79x cheaper than GPT-Sol 6.1.
Each of these models is now running in production on Microsoft’s second generation Maia 200 AI accelerator for datacenters, and MAI says it’s started making customizable versions of its models for customers with specific needs. The first is a custom model made for 5G network operation analytics that outperforms GPT-6.1 Sol at 50x cheaper costs.
MAI is also introducing MAI Licensed Weights, which lets organizations deploy and customize Microsoft AI models in their own environments on Azure. MAI Licensed Weights is coming soon through the Microsoft Frontier Company program.
“MAI is a family of world class models with exceptional price-performance: the world’s best image editing model, the bests transcription model, frontier cyber defense at 2 percent of the cost; and world class everyday coding models at a price you can truly scale,” Mr. Sulleyman added at the close of his keynote segment. “And crucially all of it with Your IQ. Your Agents. Your Weights. All under your control. This is Frontier Intelligence you can trust, from the company you can trust.”
AI 热点