Microsoft unveils in-house AI models, agents and Majorana 2 chip

Per Microsoft's Build live blog, Microsoft introduced a set of new in-house models branded as Microsoft AI, listing MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Transcribe-1.5, and MAI-Voice-2 as part of the announcement. Per Microsoft's Build live blog, the company unveiled Majorana 2, its next-generation quantum chip, and reported qubits that are 1,000x more reliable than the previous generation with a mean qubit lifetime of 20 seconds, with some instances lasting as long as one minute. Reporting by Seeking Alpha documents new agent and search products shown at Build, including an autonomous agent called Scout Autopilot and a search/agent capability labeled Web IQ. TechRadar's live coverage reported a partnership announcement with the Mayo Clinic to develop a Frontier health model, attributed to remarks in the keynote.

Per Microsoft's Build live blog, the quantum team attributed gains in Majorana 2 to improvements in topological qubit design and to using agentic AI to accelerate firmware and calibration workflows. The Build coverage lists MAI- model names but does not publish core model parameter counts or training-dataset specifics in the cited reporting. PCMag's live notes include product framing that emphasizes "agent-first" devices and developer toolchains tied to the announced models and agent capabilities.

Companies pursuing internal foundation-model families while also delivering agent orchestration and domain connectors are following an industry pattern where vertical integration aims to reduce external dependency and improve latencies and data governance. Organizations adopting autonomous agents at scale typically face higher demands for runtime observability, safety layers, and cost controls around model invocation. Significant reported improvements in qubit coherence, if independently validated, would shift the quantum developer focus from purely hardware validation toward higher-level application development and integration with classical HPC and simulation tooling.

The combination of new foundation models, agent tooling, and hardware progress highlights three concurrent vectors that matter to practitioners: model access and governance, agent orchestration and developer tooling, and emerging quantum compute capacity. Public reporting frames these announcements as part of a broader push by major cloud and platform vendors to offer full-stack AI solutions that span model, runtime, and specialized hardware. Quibit lifetime claims for Majorana 2 are unusually large relative to typical superconducting or trapped-ion numbers reported publicly, making independent validation and reproducible benchmarks critical for assessing practical impact.

For practitioners: Monitor SDK and API surface changes for the MAI- models and Scout/Web IQ agent integrations for details on latency, cost, fine-tuning or retrieval augmentation support, and enterprise security controls. For researchers and infrastructure teams: watch for published Majorana 2 performance data, reproducible benchmarks, and toolchain support that demonstrates integration of topological qubits into developer workflows. Observers should also track any follow-up documentation or technical papers Microsoft publishes that specify training datasets, model scales, and safety or evaluation results.

PCMag captured Satya Nadella saying, "It's never about tech for tech's sake," during the keynote, and TechRadar quoted Mustafa Suleyman saying, "To us, this is what owning the full stack looks like," both remarks appearing in live coverage of Build.