Microsoft (MS) has announced the launch of seven new AI models, including its first inference AI model, as part of its efforts to strengthen its independent AI ecosystem.
On June 3, during its annual developer event, Microsoft Build 2026, the company unveiled updates to its platform for developing and operating AI agents and applications.
Mustafa Suleyman, CEO of MS AI, introduced the new MAI model lineup, which includes the 'My Thinking-1' inference model. The newly announced MAI models cover various areas, including inference, coding, image editing, and voice, featuring: △Inference (My Thinking-1) △Coding (My Code-1 Flash) △Image Editing (My Image 2.5 & My Image 2.5 Flash) △Voice Transcription (My Transcribe 1.5) △Voice Generation (My Voice-2 & My Voice-2 Flash).
The My Thinking-1 model, Microsoft's first inference model, is a mid-sized model with 35 billion active parameters. It was trained from scratch using enterprise-grade commercial license data, without a distillation process from other models. It features a 256K context window and excels in executing complex multi-step instructions, long-context inference, and code generation. The model is designed for high efficiency and performance at a low token cost.
The image generation and editing model, My Image 2.5, along with its Flash version, was also revealed. My Image 2.5 is an improved version of the image generation model released by Microsoft in April, supporting both text-based image generation and image editing tasks.
The voice transcription model, My Transcribe 1.5, supports 43 languages, while the voice model, My Voice-2, offers over 15 additional languages and new voice options. The coding model, My Code-1, is optimized for GitHub and can be used with Copilot and Visual Studio Code.
Alongside the model announcements, Microsoft has also enhanced its agent development environment. The new context layer, 'Microsoft IQ,' integrates internal corporate knowledge with external information across GitHub Copilot, Microsoft Foundry, and Copilot Studio. The company also introduced a personal work agent called 'Microsoft Scout,' which understands user workflows and assists with meeting preparations, scheduling conflicts, and repetitive tasks using tools like Teams and Outlook.
Industry analysts suggest that Microsoft is shifting its strategy to reduce reliance on external AI models while simultaneously developing its own models and platforms. Unlike its previous focus on expanding generative AI services through collaboration with OpenAI, Microsoft is now clearly moving to enhance its own model and platform capabilities.
Since 2019, Microsoft has invested approximately $13 billion in OpenAI, maintaining a collaborative relationship. Microsoft has provided substantial cloud computing resources and integrated OpenAI's research outcomes into its services and cloud ecosystem. However, following the growth of ChatGPT, OpenAI has begun establishing its own infrastructure, leading to changes in the interests of both parties.
In October of last year, during OpenAI's restructuring into a profit-oriented public benefit corporation (PBC), Microsoft's stake was adjusted from 32.5% to 27%. This shift indicates that Microsoft is accelerating its own AI model development while maintaining collaboration with OpenAI, but placing greater emphasis on building an independent ecosystem. Analysts expect Microsoft to focus on enhancing cost efficiency and control in enterprise environments.
Satya Nadella, Microsoft CEO and Chairman, emphasized that the core of this announcement is not about a single technology but about supporting value creation and expansion on a platform that spans from edge to cloud in the AI stack.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.