Microsoft introduced new AI models at the Build conference, claiming they surpass competitors in performance, marking a shift towards proprietary development.
Microsoft Unveils Seven AI Models at Build Conference
Microsoft announced the release of seven new artificial intelligence models during its annual Build conference, emphasizing their superior performance compared to competitors. Specifically, the MAI-Thinking-1 model excelled in blind testing, surpassing Anthropic's Claude Sonnet 4.6 and equating to Claude Opus 4.6 in coding tests.
Additionally, the MAI-Image-2.5 model reportedly outperformed Google's image-editing technology, Nano Banana 2. This launch signifies a pivotal moment for Microsoft as it moves toward developing its own frontier AI models rather than merely supporting OpenAI initiatives.
Highlights from CEO Mustafa Suleyman
Mustafa Suleyman, CEO of Microsoft AI, expressed enthusiasm about the announcement on X, stating, "We are thrilled to unveil seven new world-class MAI models today. They signify a new era in AI that empowers users and keeps them at the forefront of technology." He highlighted MAI-Thinking-1's impressive 97% score on the AIME 2025 reasoning benchmark and praised its performance in coding assessments.
Alongside MAI-Thinking-1, Microsoft introduced various models tailored for specific functionalities: MAI-Code-1-Flash for coding optimized for GitHub Copilot, MAI-Image-2.5 and its Flash counterpart for advanced image editing, MAI Transcribe-1.5 for multilingual transcription, and MAI-Voice-2, which generates natural-sounding voice outputs in multiple languages.
Suleyman noted the exponential growth in computational power devoted to training AI models, predicting significant advancements in capabilities over the next few years. His remarks coincide with increasing competition in the AI sector, with Anthropic and Google recently announcing their own developments.
Strategic Shift Towards Proprietary Technology
This latest model release marks a strategic pivot for Microsoft as it seeks to establish itself as a primary AI innovator, asserting that its MAI models deliver superior quality at significantly lower costs compared to GPT-5.5. This aligns with a broader industry trend toward creating AI that is more controllable and responsive to developer and business needs.
