Microsoft announced two major technological breakthroughs at Build 2026 on June 2, 2026: the Majorana 2 quantum chip with dramatically improved qubit stability and MAI-Thinking-1, its first in-house reasoning model trained entirely without OpenAI data.
Majorana 2 Achieves 1,000-Fold Reliability Improvement
The Majorana 2 quantum chip represents a significant advance over its predecessor, Majorana 1, which was announced February 19, 2025. The new chip delivers substantial improvements in qubit stability:
- 1,000-fold improvement in reliability compared to previous generations
- Average qubit lifetimes extended to approximately 20 seconds
- Some qubits maintaining stability for as long as one minute
- Transition from aluminum-based to lead-based superconducting structures
- Enhanced resistance to environmental disturbances including cosmic rays and background radiation
Microsoft developed the chip with assistance from Microsoft Discovery, its agentic AI research platform that uses autonomous agents to accelerate complex research. The company has cut its quantum computing timeline in half and now aims to deliver a scalable, practical quantum computer by 2029.
However, the announcement has drawn skepticism from some experts. Physicist Henry Legg of the University of St Andrews stated: "Nothing in this preprint resolves the fundamental issues."
MAI-Thinking-1 Debuts as First In-House Reasoning Model
Microsoft's AI Superintelligence Team released a family of seven new models, with MAI-Thinking-1 as the flagship reasoning model. The 35 billion active parameter model features:
- 256K context window for processing extensive information
- Mixture-of-Experts (MoE) architecture that selectively activates subcomponents per request
- Training from scratch with zero distillation on enterprise-grade, commercially licensed data
- No OpenAI data used in training
The model demonstrates competitive performance on industry benchmarks:
- Independent raters prefer it to Claude Sonnet 4.6 in blind tests
- Matches Claude Opus 4.6 on coding abilities on SWE Bench Pro
- 97.0% accuracy on AIME 2025
- 94.5% accuracy on AIME 2026 mathematical and multi-step scientific reasoning benchmarks
MAI-Thinking-1 excels at complex multi-step instructions, long-context reasoning, code generation, function calling, and multi-layered instruction following. The model is compatible with the Chat Completions API and available on Microsoft Foundry in private preview.
Additional AI Announcements
Microsoft also announced MAI-Image-2.5 for text-to-image and image-to-image generation, and MAI-Transcribe-1.5 featuring state-of-the-art accuracy with entity biasing and streaming capabilities. The company announced general availability of Microsoft Discovery, its agentic AI research platform.
Key Takeaways
- Microsoft's Majorana 2 quantum chip delivers a 1,000-fold reliability improvement with qubit lifetimes reaching 20 seconds on average and up to one minute for some qubits
- The company has accelerated its quantum computing roadmap, now targeting delivery of a practical quantum computer by 2029
- MAI-Thinking-1, a 35 billion parameter reasoning model with 256K context window, was trained entirely without OpenAI data and matches Claude Opus 4.6 on coding benchmarks
- The model achieved 97.0% on AIME 2025 and 94.5% on AIME 2026 mathematical reasoning benchmarks
- Microsoft announced a family of seven new in-house AI models including MAI-Image-2.5 and MAI-Transcribe-1.5, with general availability of its Microsoft Discovery agentic research platform