Microsoft's MAI-Thinking-1: AI for Legal Compliance

Microsoft is launching a family of seven new AI models, with a major focus on legal compliance for businesses. The star of the show is MAI-Thinking-1, a 35-billion-parameter model that Microsoft calls its first "reasoning model." What's interesting is that MAI-Thinking-1 was trained "entirely from the bottom," without using other AI models. Microsoft is pitching this to businesses by highlighting the legal concerns around data sourcing and copyright infringement in AI. They say their model has "enterprise-grade, clean, and commercially licensed data lineage," offering trustworthiness. Other new models include MAI-Image-2.5 for image generation, MAI-Transcribe-1.5, which is described as the "best transcription model in the world," and speech-generation models MAI-Voice-2 and Mai-Voice-2 Flash. There's also MAI-Code-1-Flash for generating code. The bottom line: Microsoft is making a big push with its own in-house AI, emphasizing legal safety to attract enterprise customers.