Microsoft Expands MAI AI Models With New Reasoning and Coding Systems at Build 2026

Microsoft has expanded its in-house MAI AI model family during Build 2026, introducing several new models focused on reasoning, coding, speech, and image generation capabilities, according to Neowin.

The announcement includes Microsoft’s first reasoning-focused AI model, a new coding model for GitHub Copilot, upgraded speech systems, and broader availability for its latest image-generation technology.

Microsoft also unveiled Microsoft Scout, the Rayfin platform for AI applications, and the new Intelligent Terminal experience during the event.

Microsoft Continues Expanding the MAI Model Family

Microsoft has steadily expanded its MAI lineup over the past year as it builds more in-house AI capabilities.

Previous MAI releases include:

• MAI-Voice-1
• MAI-1-preview
• MAI-Transcribe-1
• MAI-Image-2
• MAI-Image-2.5

The latest Build 2026 announcements significantly broaden the portfolio with new models aimed at enterprise AI workloads, coding assistance, speech processing, and multimodal experiences.

MAI-Thinking-1 Becomes Microsoft’s First Reasoning Model

One of the biggest announcements is MAI-Thinking-1, Microsoft AI’s first dedicated reasoning model.

According to Microsoft, the model was trained entirely from scratch without using distillation techniques. The company says it relied on clean, commercially licensed, enterprise-grade training data during development.

MAI-Thinking-1 features 35 billion active parameters and supports a 128K context window, allowing it to process large prompts and extended conversations.

Built for Complex Reasoning and Coding Tasks

Microsoft says MAI-Thinking-1 is designed for complex multi-step instructions, long-context reasoning, and code generation tasks.

The company positions the model for workloads that require deeper reasoning and advanced problem-solving capabilities.

Benchmark Claims

Microsoft says blind testers preferred MAI-Thinking-1 over Claude Sonnet 4.6 in comparative evaluations.

The company also claims the model matches Claude Opus 4.6 on coding tasks in SWE-bench Pro.

However, Microsoft did not release extensive benchmark comparisons against competing frontier AI models.

MAI-Thinking-1 is currently available in private preview through Microsoft Foundry.

MAI-Image-2.5 Expands Across Microsoft Products

Microsoft also announced broader availability for MAI-Image-2.5 alongside a new flash variant for developers.

The company says the updated image-generation model improves text rendering, stylized illustrations, and commercial imagery generation.

MAI-Image-2.5 is already integrated into PowerPoint and is currently rolling out to OneDrive. Developers can access the model through Microsoft Foundry.

Microsoft Introduces MAI-Transcribe-1.5

Microsoft unveiled MAI-Transcribe-1.5, an upgraded speech-to-text model.

The new version supports 43 languages and, according to Microsoft, delivers state-of-the-art transcription accuracy across supported languages and accents.

Streaming transcription support is expected to arrive soon.

MAI-Voice-2 Adds More Languages and Voices

The company also introduced MAI-Voice-2 and a flash variant.

The updated speech-generation model adds support for more than 15 additional languages and includes new voice options for developers and enterprise customers.

MAI-Code-1 Powers GitHub Copilot Experiences

Another major announcement is MAI-Code-1, Microsoft’s new coding-focused AI model.

The model is optimized for GitHub-related workloads and designed to remain inference-efficient while handling developer tasks.

MAI-Code-1 is now available in GitHub Copilot and Visual Studio Code.

MAI Models Expand to Third-Party Platforms

Microsoft also confirmed that MAI models will become available through several third-party AI platforms, including Fireworks AI, Baseten, and OpenRouter.

In addition, Fireworks AI is now generally available inside Microsoft Foundry.

The broader rollout suggests Microsoft wants MAI models to compete more directly with other commercial AI platforms while expanding availability beyond Microsoft’s own ecosystem.