Microsoft Unveils Seven New Models at Build: Flagship Reasoning Model Challenges Anthropic, Creating a "Thinking + Coding" Agent Loop

Microsoft has begun to directly challenge Anthropic's stronghold.

At the annual Build developer conference held on Tuesday, the 2nd (US Eastern Time), Microsoft unveiled seven new AI models in one go, covering multiple directions including reasoning, coding, vision, and multimodality. The most notable among them are Microsoft's first flagship reasoning model—MAI Thinking-1—and MAI-Code-1-Flash, a coding model focused on GitHub scenarios that emphasizes low cost and high efficiency.

Mustafa Suleyman, head of Microsoft AI, stated in an interview with media outlets that Microsoft is pursuing a development path different from Google, Meta, and OpenAI, focusing more on the "Anthropic-style" enterprise, developer, and programming markets. He said, "We are more focused on the Anthropic-style direction—enterprise, developers, and coding."

This week's series of model releases also marks a new stage in Microsoft's advancement of its AI "autonomy" strategy: while continuing its deep cooperation with OpenAI, Microsoft is accelerating the creation of its own cutting-edge model system.

Microsoft's Seven New Models Span from Reasoning to Coding

According to Microsoft's announcement, the seven new models released this time come from the expanded MAI (Microsoft AI) family, covering different capability levels and use cases.

Core products include:

• MAI Thinking Series Reasoning Models
• Ultra-Efficient Code Models
• Vision and Multimodal Models
• Lightweight Models for Agent Systems
• Enterprise and Developer Optimized Models

Microsoft describes this as part of building a "hill-climbing machine"—a model development system that continuously iterates and self-optimizes.

Microsoft stated that the goal of the new model portfolio is not merely to pursue parameter scale, but to build a "thinking, reasoning, execution, coding" capability stack that can support next-generation AI Agent systems.

MAI Thinking Targets Anthropic Claude Sonnet 4.6

The most significant release this time is Microsoft's debut of its reasoning model family—MAI Thinking.

Reasoning models have become the focal point of AI competition in 2026. Compared to traditional chat models, these models emphasize: multi-step deduction; complex task decomposition; long-chain planning; mathematical and code reasoning; and Agent task execution.

Microsoft stated that MAI Thinking can break down complex problems into smaller, more manageable steps, specifically optimized for coding, developer workflows, and Agent tasks.

According to data released by Microsoft, models in the MAI-Thinking series achieve coding capabilities comparable to Anthropic's Claude Sonnet 4.6, released in February this year.

Microsoft's announcement stated:

"MAI-Thinking-1 is the flagship reasoning model under Microsoft AI. As a mid-sized model, it ranks among the top in its class: its performance is on par with leading models in key software engineering benchmarks; and in blind comparative evaluations, it reached parity with Sonnet 4.6 in terms of human preference."

Mustafa Suleyman admitted to the media that Anthropic currently remains several months ahead, but emphasized that Microsoft is rapidly closing the gap:

"We are now absolutely at the forefront." "In six months, we have closed a huge gap."

This statement somewhat reflects a change in Microsoft's current AI R&D strategy—no longer satisfied with simply calling OpenAI models, it hopes to enter the front line of model competition.

Ultra-Efficient Coding Model: Microsoft Targets GitHub and Enterprise Development Markets

In addition to reasoning models, Microsoft also released an "ultra efficient" coding model, MAI-Code-1-Flash. Reportedly, this model has been fine-tuned specifically for Microsoft's GitHub developer platform. Microsoft introduced:

"MAI-Code-1-Flash is an inference-efficient agentic programming model. Designed and deeply integrated with GitHub Copilot, VS Code, and the Microsoft technology stack, its performance rivals Haiku with a 5 billion parameter scale, yet at a lower cost."

w Coding models have become one of the clearest directions for AI commercialization. Current competitors include: Anthropic Claude Code / Cowork, OpenAI's Codex system, Google's Gemini Code capabilities, and AI-native development platforms such as Cursor and Replit.

Microsoft clearly hopes to leverage the GitHub ecosystem to reinforce its advantages.

Suleyman believes that the combination of reasoning models and coding models will be key to the next stage of AI Agent development. He said that the "thinking + coding" capability combination can help Microsoft build true Agent systems—agents capable of autonomously completing tasks.

Why Is Microsoft Targeting Anthropic?

Notably, Microsoft executives publicly positioned Anthropic as the core benchmark this time.

Suleyman explicitly told the Financial Times that Microsoft is "not that concerned" about the consumer-oriented routes favored by Google, Meta, and OpenAI.

Microsoft is more focused on the enterprise developer market championed by Anthropic. The reasons are not complex.

Anthropic has risen rapidly over the past year, with its advantages increasingly concentrated in Microsoft's core commercial stronghold: enterprise software, AI programming, developer tools, and white-collar office automation.

In particular, Anthropic launched Cowork, an AI coding and office tool for enterprise users, which once triggered market concerns about the enterprise software industry. Related products even drove a collective decline in software stocks, putting pressure on Microsoft's share price, which has fallen more than 6% year-to-date, significantly underperforming the broader market, while the S&P 500 index has risen more than 10% this year.

For Microsoft, this is no longer just a competition of model capabilities, but a defensive battle for the moats of Office, GitHub, Copilot, and enterprise software.

Microsoft Accelerates Efforts to Reduce Dependence on OpenAI

At a deeper level, behind this model release is Microsoft's ongoing AI autonomy strategy.

In recent years, many of Microsoft's AI products have heavily relied on OpenAI models. From Copilot to Azure AI, and even enterprise services, OpenAI has been almost the core source of Microsoft's AI capabilities. But this landscape is changing.

Media outlets pointed out that after readjusting its partnership with OpenAI last year, Microsoft began to actively promote a route of "true self-sufficiency."

Microsoft still holds approximately 27% of OpenAI's shares and retains long-term access to advanced models, but the company has clearly begun to build a "de-single-dependency" multi-model strategy.

Over the past year, Microsoft has taken a series of actions: establishing a superintelligence research team; strengthening internal model R&D; investing in Anthropic cloud cooperation; establishing the MAI model system; and advancing agent frameworks and infrastructure.

Suleyman stated that owning self-developed models will also bring direct financial benefits.

Currently, when providing certain AI services to customers, Microsoft needs to pay partners a "significant profit share." Once internal models mature, they can significantly reduce cost pressures. Suleyman said, "This will be directly reflected on the income statement."

Risk Warning and Disclaimer

The market involves risks, and investment requires caution. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this content is at the user's own risk.