Microsoft The first model of reasoning is one of seven AIs just released at the Build General Assembly.

Microsoft AI has introduced several new models on the build.

One of them is the first argumentation model of the company.

With Nano Banana Pro a new image model appears competitive.

On Tuesday, Microsoft opened its annual build developer conference with a keynote where the company announced seven new AI models, including its first argumentation model. During the keynote, Mustafa Suleyman, CEO of Microsoft AI, reaffirmed the laboratory’s “humanistic superintelligence” strategy during the presentation of the new models.

The first argumentation model of Microsoft AI, MAI-Thinking-1, was trained on “business-ready, clean and commercially licensed data”, said the company in the blog announcement. Given the increasing concerns (and active actions) about copyrights and the use of AI, it will be important for Microsoft customers to address this, but it is not the first company to make such a promise.

Microsoft said that the 35 billion parameter model Anthropics Sonnet 4.61 surpassed when it was evaluated by independent reviewers in a blind test, and that it matches Anthropic Opus 4.6 in its SWE Bench Pro benchmark score for coding. In line with the Agent-Wahn, which currently records all AI labs, MAI-Thinking-1 is designed for multi-stage tasks and is now available in Microsoft Foundry's private preview.

The Microsoft AI family (and the general race for the development of the best coding model on the market) also includes MAI code-1, which the company is “ultraefficient” and “optimized for GitHub referred to. MAI-Code-1 comes to Copilot and VS Code on Tuesday.

New language and image models

In addition to its flash equivalent, MAI-Image-2.5 is the first model of Microsoft for text-to-image and image-to-image tasks. According to the company, Nano Banana Pro surpassed ELO, an assessment system acquired from chess that measures relative skills. The MAI image 2.5 models are now available in PowerPoint and Foundry and are introduced into OneDrive. As Suleyman announced during the keynote, there had already reached the third place in the LM Arena Leaderboard, just behind Nano Banana.

MAI-Transcribe-1.5 “combines state-of-the-art accuracy in 43 languages, streaming is soon available,” said Microsoft. The company also published MAI-Voice-2 and its Flash-Sidekick, which is available in 15 languages more than its predecessor MAI-Voice-1. Microsoft has only published the previous generations of these models in the preview two months ago, thus showing once again how quickly the introduction cycle for new AI models has become this year.

“Everything is provided with a watermark from the ground,” Suleyman stressed and pointed to the safety frames of the new models. He also mentioned cost-efficiency improvements in each model, some of which range up to ten times compared to similar competition models. All new MAI models will be available on Fireworks AI, which is now generally available on Foundry, according to the company, as well as on Baseten and Open Router.

A new medical partnership

Suleyman concluded his announcement on the AI model with the announcement of a collaboration with the Mayo Clinic to develop a new border model for healthcare. The project joins the growing number of health-specific AI applications from companies, including OpenAI and Google. Microsoft already offers Copilot Health. Nevertheless, data protection, security and hallucinations give cause to worry when it comes to medical AI.