Microsoft Unveils Wave Of AI Tools, Platforms, And Quantum Breakthrough At Build 2026
Microsoft used its Build 2026 developer conference in San Francisco to announce a sweeping set of artificial intelligence products, platforms, and a major quantum computing advance, signaling the company’s most aggressive push yet into the agentic AI era. The announcements spanned in-house AI models, new search infrastructure for AI agents, an AI behavior testing framework, a next-generation device platform, an always-on personal work agent, and a next-generation quantum chip, collectively painting a picture of a company repositioning itself from a software provider to an AI-native enterprise.
Leading the model announcements was MAI-Thinking-1, Microsoft’s first in-house reasoning model and the flagship offering from its AI Superintelligence team. The mid-sized model carries 35 billion active parameters and a 128,000-token context window, and was trained from scratch on clean, commercially licensed data without distilling outputs from third-party models. Microsoft said independent evaluators preferred MAI-Thinking-1 over competing models in blind testing, and that it matches Claude Opus 4.6 on coding benchmarks in SWE-bench Pro. The model is currently available in private preview through Microsoft Foundry. Alongside it, Microsoft unveiled MAI-Code-1 for GitHub Copilot and VS Code, MAI-Image-2.5 for PowerPoint and OneDrive, and MAI-Transcribe-1.5, a transcription model the company says achieves state-of-the-art accuracy across 43 languages.
On the infrastructure side, Microsoft launched Web IQ, a new grounding API suite built on Bing’s index but re-engineered specifically for how AI agents search. Unlike traditional search, which prioritizes ranked results for human users, Web IQ is designed to extract and package precise information from web documents as efficiently as possible for AI systems during inference. Microsoft said the system is roughly 2.5 times faster than the next best alternative and already powers both Microsoft Copilot and OpenAI’s ChatGPT. Microsoft also unveiled ASSERT, short for Adaptive Spec-driven Scoring for Evaluation and Regression Testing, an open-source framework that lets developers generate AI behavior tests from plain-language descriptions. The tool converts high-level policy descriptions into structured test cases, runs them against a target AI system, scores the results, and logs the paths taken by the AI so developers can identify where failures occur. Rounding out the software announcements were Project Solara, an Android-based platform for agent-first devices with two reference hardware designs, including a desktop hub and a wearable employee badge, and Scout, an always-on agent integrated across Microsoft 365 apps, including Teams, Outlook, OneDrive, and SharePoint. Early Solara pilots are lined up at AccuWeather, Best Buy, CVS Healthcare, and Target.
Closing out the conference was the unveiling of Majorana 2, Microsoft’s next-generation topological quantum chip developed with assistance from its Discovery agentic AI platform. The chip replaces aluminum with lead as its superconductor and updates the semiconductor active region with a new materials stack, delivering a 1,000-fold improvement in qubit reliability over the previous generation. Majorana 2 qubits now have a mean lifetime of 20 seconds, with some instances sustaining quantum states for up to one minute, compared to the millisecond-range lifetimes of earlier designs. On the basis of that progress, Microsoft said it has cut its quantum computing roadmap in half and now targets a scalable, commercially useful quantum computer by 2029.
