Microsoft Challenges OpenAI with Seven Innovative AI Models

Microsoft Strengthens Its AI Strategy with Seven New Models

On June 2, 2026, during its annual Build conference, Microsoft announced the launch of seven new artificial intelligence models, marking a significant step towards independence from OpenAI. These models, grouped under the MAI banner, aim to directly compete with offerings from OpenAI, Anthropic, and Google across various technological domains.

A New Era for Microsoft with MAI

Historically, Microsoft has closely collaborated with OpenAI, investing several billion dollars in the company. However, recent adjustments to their partnership, particularly in October 2025 and April 2026, have allowed Microsoft to develop its own models. These models are fully trained from scratch, without distillation from third-party models, on commercially licensed data, a major asset for companies concerned about copyright issues.

Among these new models, MAI-Thinking-1 stands out particularly. Equipped with 35 billion parameters and utilizing a MoE architecture, this reasoning model is classified among medium-sized models. Independent tests have demonstrated its superiority over Claude Sonnet 4.6 from Anthropic, and its performance on the SWE Bench Pro benchmark brings it close to Claude Opus 4.6.

A Diverse Multimodal Range

Microsoft's seven new models cover a wide range of functionalities:

• MAI-Thinking-1: Designed for complex tasks, this model is currently in private preview on Microsoft Foundry.

• MAI-Code-1-Flash: A code model optimized for VS Code and GitHub Copilot, with 5 billion parameters, already deployed in these environments.

• MAI-Image-2.5: This image generation and editing model claims the 2nd place in the Arena ranking for image editing, just behind Nano Banana 2. It is integrated into PowerPoint and is being deployed on OneDrive.

• MAI-Image-2.5-Flash: A lightweight version of MAI-Image-2.5, designed for high production loads.

• MAI-Transcribe-1.5: Capable of transcribing 43 languages, this model is five times faster than its competitors and is integrated into GitHub, Teams, Copilot, and Dynamics 365 Contact Center.

• MAI-Voice-2: A speech synthesis model available in 15 languages, offering precise emotional control.

• MAI-Voice-2-Flash: A low-latency variant of MAI-Voice-2, ideal for voice agents requiring quick responses.

All these models are available on Microsoft Foundry and accessible via OpenRouter, Fireworks, and Baseten, allowing developers to customize them according to their needs.