Microsoft unveils seven AI models, claims wins over Claude and Google
In brief
- Microsoft introduced MAI-Thinking-1 reasoning model, preferred over Claude Sonnet 4.6 in blind tests.
- MAI-Image-2.5 and MAI-Code-1-Flash join transcription and voice models spanning 43 and 15 languages.
- Microsoft claims 10x lower cost than competing systems while delivering superior performance.
Flagship reasoning model leads the lineup
Microsoft's MAI-Thinking-1, the company's flagship text foundation model, was preferred over Anthropic's Claude Sonnet 4.6 in blind tests conducted by independent evaluators. The model scored 97% on AIME 2025, a benchmark measuring advanced problem-solving and reasoning skills.
Microsoft said MAI delivered the highest win rate, outperforming GPT-5.5 on quality while being 10x lower on cost. The cost claim underscores the company's broader strategy to compete on both performance and efficiency as AI development accelerates.
Specialized models across coding, vision, and audio
The company introduced MAI-Code-1-Flash, a lightweight coding model built for GitHub Copilot and Visual Studio Code. On the vision side, MAI-Image-2.5 and its Flash variant outperform Google's Nano Banana Pro on image-editing tasks.
Microsoft also rolled out transcription and voice capabilities. MAI Transcribe-1.5 supports 43 languages, while MAI-Voice-2 produces natural-sounding speech in 15 languages. The breadth of the lineup suggests Microsoft intends to compete across multiple AI verticals rather than focusing on a single flagship.
Competitive context
Anthropic announced its latest flagship model, Opus 4.8, which the company said is faster and smarter on benchmark tests. Google unveiled Gemini Omni, a multimodal AI model combining Gemini with Veo, Nano Banana, and Genie media-generation models, alongside Gemini Spark, a cloud-based AI agent for task management across apps and workflows.
Microsoft's push into frontier model development reflects broader industry momentum. The company remains OpenAI's largest backer and infrastructure partner, but these new models show it's investing heavily in proprietary AI research and deployment.
Frequently asked questions
What is MAI-Thinking-1 and how does it compare to Claude?
MAI-Thinking-1 is Microsoft's flagship text foundation model. In blind tests conducted by independent evaluators, it was preferred over Anthropic's Claude Sonnet 4.6. The model scored 97% on AIME 2025, a benchmark measuring advanced problem-solving and reasoning.
How many languages does Microsoft's transcription and voice models support?
MAI Transcribe-1.5 supports 43 languages for transcription, while MAI-Voice-2 generates natural-sounding speech in 15 languages. Together, these models expand Microsoft's AI capabilities across global audio and speech workflows.
Why is Microsoft launching its own AI models?
Microsoft is attempting to establish itself as a frontier AI developer rather than solely relying on its role as OpenAI's largest backer and infrastructure provider. The seven-model lineup demonstrates the company's investment in proprietary AI research and deployment across multiple verticals.
