Microsoft kicked off Microsoft Build 2026 at Fort Mason Center in San Francisco on Tuesday, June 2, unveiling seven in-house AI models led by MAI-Thinking-1 — the company's first reasoning model, built from scratch on commercially licensed enterprise data with no distillation from third-party models, including OpenAI's GPT series. The keynote, delivered by Satya Nadella beginning at 9:30 a.m. PT, marked the most concrete evidence yet that Microsoft's multi-vendor, multi-model strategy has moved from roadmap to shipping product.
MAI-Thinking-1: Microsoft's First Reasoning Model Enters Private Preview
MAI-Thinking-1 is a mid-sized model with 35 billion active parameters and approximately one trillion total parameters in a sparse Mixture of Experts architecture, along with a 256,000-token context window — enough, Microsoft said, to process a 600-page document in a single pass. Microsoft trained it entirely on clean, commercially licensed data without distillation from any third-party model, which the company describes as a key enterprise differentiator because it gives customers confidence about the provenance of what the model learned.
The model reaches 97.0 percent on AIME 2025 and 94.5 percent on AIME 2026, benchmarks that test mathematical and multi-step scientific reasoning. On SWE-Bench Pro — a software engineering benchmark — Microsoft says it matches Claude Opus 4.6 on coding tasks, and in blind side-by-side evaluations run by Surge, Microsoft's independent human rating partner, MAI-Thinking-1 was preferred over Claude Sonnet 4.6. Microsoft has published a preprint describing the evaluation methodology, but full reproduction of the results by independent labs has not yet occurred, leaving those benchmark claims open to challenge until confirmed externally.
MAI-Thinking-1 is now available in private preview through Microsoft Foundry. It supports function calling, multi-layered instruction following, and is compatible with the widely used Chat Completions API.
MAI-Code-1-Flash Rolls Out to All GitHub Copilot Plans
Alongside the reasoning model, Microsoft confirmed that MAI-Code-1-Flash — a small-tier inference-efficient coding model — is rolling out today across all GitHub Copilot tiers, including Free, Pro, Pro+, and Max plans, beginning with a limited set of users and expanding gradually over the coming weeks. The model was trained inside GitHub Copilot's production harness rather than benchmarked externally and then deployed into it, a distinction Microsoft says improves reliability in agentic coding workflows.
Microsoft has not publicly connected MAI-Code-1-Flash to any specific migration timeline for existing Copilot defaults. Some post-keynote coverage described the model under the name "Project Polaris" and reported a plan to replace GPT-4 Turbo as the default Copilot engine in August 2026; those details do not appear in Microsoft's official Build 2026 announcement or in published GitHub Copilot release documentation as of this writing and should be treated as unverified reporting until Microsoft publishes formal product documentation.
Five More MAI Models Span Voice, Transcription, and Imaging
The full seven-model launch covers the breadth of Microsoft's developer surface. MAI-Image-2.5 and its flash variant are Microsoft's first models handling both text-to-image and image-to-image workloads, ranking third and second respectively on the Arena AI leaderboard. MAI-Image-2.5 is already live in PowerPoint and rolling out to OneDrive. MAI Transcribe 1.5 reaches state-of-the-art accuracy across 43 languages, with streaming support coming soon and a claimed five-times speed improvement over competing transcription models. MAI-Voice-2 brings high-quality speech generation across more than 15 additional languages, with voice adaptation from a short sample; its flash variant is coming soon.
All MAI models are also available outside Microsoft's own catalog: Fireworks AI — now generally available on Foundry — provides a single platform experience with enterprise governance and Azure data residency regardless of model choice, and MAI models will additionally be distributed through Baseten and OpenRouter.
Is MAI-Code-1 Replacing GPT-4 in GitHub Copilot?
Not exactly, and not yet. What Microsoft confirmed at Build 2026 is that MAI-Code-1-Flash is now rolling out to Copilot users — beginning gradually — as an option in the Copilot model picker. The official announcement does not describe it as a replacement for any existing model, and GitHub Copilot continues to support multiple model backends including OpenAI's. Any specific migration timeline, such as an August 2026 default switch from GPT-4 Turbo to an in-house model, has not been confirmed in Microsoft's own published materials.
The context for that question is real: Microsoft and OpenAI amended their partnership in April 2026, ending Microsoft's exclusive license to OpenAI intellectual property and removing Microsoft's revenue share obligation to OpenAI, while preserving OpenAI's capped revenue share to Microsoft through 2030. The renegotiation, which briefly sent Microsoft shares down before a recovery, means Microsoft is no longer contractually tied to OpenAI as its sole model provider. William Blair analyst Jason Ader summarized the revised terms at the time: securing IP rights through 2032 protects the foundation of Microsoft's Copilot strategy and Azure OpenAI monetization, but Azure will now have to compete more actively for OpenAI workloads. NYU Stern professor Robert Seamans described the dynamic as Microsoft "continuing to rely on a really, really important partner, but also hedging their bets."
Claude Remains in Azure AI Foundry, Running on Anthropic Infrastructure
Claude models — now including Claude Opus 4.8 (released May 28, 2026), Opus 4.7, Opus 4.6, Sonnet 4.6, and Haiku 4.5 — remain available in Microsoft Foundry under unified Azure billing, eligible for Microsoft Azure Consumption Commitment (MACC), and positioned as partner-tier frontier models alongside OpenAI offerings. Nothing at Build 2026 changed that arrangement.
One operational distinction deserves attention before any enterprise team treats Claude in Foundry as structurally equivalent to OpenAI models on Azure: Claude models currently run on Anthropic-managed infrastructure, not on Azure's regional compute. That is a different operational profile from OpenAI models, which run natively on Azure. A Microsoft Q&A response from April 2026 confirmed the gap and noted that EU-region native inference support for Claude in Foundry is listed as "Coming 2026" on Anthropic's regional compliance page, with no more specific date publicly announced. For enterprises with strict data residency requirements, particularly in the European Union, that split-infrastructure arrangement may require supplementary compliance planning.
What the Anthropic relationship represents in the Build 2026 context is the broader thesis: Microsoft is not asking enterprise developers to choose between OpenAI and Anthropic. It is building Foundry as the orchestration layer above all of those choices — unified billing, unified governance, and model-routing flexibility, regardless of which frontier lab built the model.
Majorana 2 Quantum Chip: Big Claims, Pending Peer Review
Microsoft closed its Build keynote with Majorana 2, a next-generation quantum computing chip that the company says achieves an average qubit lifetime of 20 seconds with individual instances reaching up to one minute — 1,000 times higher reliability than its previous generation. Microsoft says the design puts it on a path to one million qubits on a chip that fits in the palm of a hand and sets a target of a commercially valuable scalable quantum machine by 2029.
Independent physicists are skeptical. Scientific American reported that outside experts say the technology "doesn't even work and never has." Critics note that the Majorana 2 results rest on a preprint manuscript that has not yet been peer-reviewed, and that the paper presents only Z measurements — not both X and Z measurements required to demonstrate a functioning topological qubit. Henry Legg, a physicist at the University of St. Andrews, said the data in the preprint come from a handful of instances on a single device and that the result "would never make it through peer review" by any other group. Sergey Frolov of the University of Pittsburgh noted that Microsoft's last preprint of this kind has remained unpublished since last summer, which he cited as evidence that journals have likely rejected it. Microsoft's history with this technology adds context: a 2018 claim to have observed Majorana zero modes was retracted after independent scrutiny, and Majorana 1 (introduced in 2025) also faced expert criticism before Majorana 2's announcement. Some industry figures responded more positively, with Terra Quantum CEO Markus Pflitsch calling it "truly an advance for the industry." The 2029 target will require the same independent peer review that the current preprint has not yet received.
Frontier Tuning and Agent Infrastructure Round Out the Build Stack
Beyond models, Microsoft announced Frontier Tuning — now in private preview — which applies reinforcement learning within a customer's compliance boundary so agents can learn an organization's specific workflows, domain knowledge, and data without exporting anything outside the enterprise perimeter. The approach is designed to let agent systems sharpen over time without violating data governance requirements.
The Microsoft Discovery platform reached general availability at Build 2026, providing an enterprise-grade agentic AI platform for scientific research workflows. BHP is using it to find copper-leaching solutions in months instead of years; Syensqo is accelerating semiconductor research; GSK is iterating on drug discovery. A free Discovery local app — available in preview with only a GitHub Copilot account required — was announced for the broader scientific community.
What the Build 2026 Announcements Mean for Enterprise Developers
The strategic read on Build 2026 is not that Microsoft has severed or degraded its OpenAI relationship. Azure remains OpenAI's primary infrastructure. GitHub Copilot continues to support OpenAI models. Microsoft 365 Copilot continues to use OpenAI capabilities. None of that changed on June 2.
What changed is that Microsoft now has operational, shipping alternatives at every tier of the developer stack: an in-house reasoning model, a coding model tuned to production Copilot workflows, transcription and voice models across dozens of languages, and image generation capabilities running in PowerPoint and OneDrive. The strategy is multi-model, multi-vendor, and designed to let enterprise teams route across providers without leaving Azure's billing and governance perimeter.
Enterprise teams evaluating that perimeter should note that the Federal Trade Commission has an active and expanding antitrust investigation into Microsoft's cloud and AI bundling practices, with civil investigative demands issued to at least six competing companies, including AWS and Google Cloud. The probe examines whether Microsoft uses its dominance in productivity software to lock customers into Azure. The UK's Competition and Markets Authority has opened a parallel Strategic Market Status investigation into Microsoft's business software ecosystem, with a designation decision expected by February 2027. Neither investigation has resulted in enforcement action, but both are ongoing and relevant to any enterprise team's vendor-diversification strategy.
The cleaner message from Build 2026 is that Microsoft is standardizing on Foundry and Copilot as the orchestration layer above all model choices. The most consequential announcements are the ones that make that layer more credible: official documentation, formal product surfaces, and the measured language Microsoft's own blog used on June 2 — where every claim about MAI-Thinking-1, MAI-Code-1-Flash, and the broader agent stack was tied to a specific product surface, not a projection.
Frequently Asked Questions
What is MAI-Thinking-1?
MAI-Thinking-1 is Microsoft's first in-house reasoning model, announced at Build 2026 on June 2, 2026. It is a mid-sized sparse Mixture of Experts model with 35 billion active parameters and a 256,000-token context window, trained entirely on commercially licensed data without distillation from any third-party model. It is available now in private preview on Microsoft Foundry.
What is MAI-Code-1, and is it replacing GPT-4 in GitHub Copilot?
MAI-Code-1-Flash is Microsoft's inference-efficient coding model tuned specifically for GitHub Copilot workflows. It began rolling out to all Copilot tiers — Free, Pro, Pro+, and Max — on June 2, 2026, starting with a limited set of users and expanding gradually. Microsoft has not published a confirmed timeline for replacing any specific default model; reports of an August 2026 switch from GPT-4 Turbo appear in post-keynote coverage but are not confirmed in official Microsoft documentation as of this writing.
What did Microsoft announce about the Majorana 2 quantum chip at Build 2026?
Microsoft announced Majorana 2, a next-generation quantum computing chip with qubit lifetimes averaging 20 seconds and a claimed 1,000-times reliability improvement over its previous chip. Microsoft targets a commercially valuable scalable quantum machine by 2029. The announcement is based on a preprint that has not yet been peer-reviewed, and several independent physicists have publicly challenged the underlying claims about topological qubits, with one researcher stating the results would not survive standard journal peer review.
Does Claude have full Azure infrastructure parity with OpenAI models in Microsoft Foundry?
No. Claude models in Microsoft Foundry currently run on Anthropic-managed infrastructure, not on Azure's regional compute infrastructure. This means EU data residency requirements may require supplementary planning; Anthropic lists EU-native inference support in Foundry as "Coming 2026" with no more specific date. Claude is billed through Azure and eligible for Microsoft Azure Consumption Commitment (MACC), but its operational profile differs from OpenAI models, which run natively on Azure.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
