# 微軟發表首款大型語言模型 MAI-Thinking-1，據稱其性能可媲美 Anthropic 的 Claude Opus 4.6。

*genai · news · 2026-06-03 · Legal IT Insider*

## Key points

- 微軟發表首款大型語言模型 MAI-Thinking-1，可與 Anthropic 的 Claude Opus 4.6 競爭。
- MAI-Thinking-1 及更新模型將驅動 Copilot、Bing、PowerPoint 及 Azure Speech。
- 微軟推出 MAI-Code-1-Flash，能從書面描述生成應用程式代碼。
- 這些模型標誌著微軟從依賴 OpenAI 轉向自主擁有及商業化 AI 的轉變。
- 微軟的模型多元化策略旨在降低成本並深化治理與安全的整合。

Microsoft has unveiled its first large language model called MAI-Thinking-1, which it says matches the performance of Anthropic’s Claude Opus 4.6 and which underpins its efforts to build frontier AI free from OpenAI. The tech giant at its Microsoft Build conference in San Francisco yesterday (2 June) also unveiled an updated image generation model (MAI-Image-2.5); an updated multilingual text-to-speech model (MAI-Voice-2); and an updated speech-to-text model that supports 43 languages (MAI-Transcribe-1.5.). Together with MA1-Thinking-1, these are the same models powering Copilot, Bing, PowerPoint and Azure Speech, will all be available in Foundry for developers to build with. And Microsoft unveiled MAI-Code-1-Flash, which takes written descriptions and generates the source code for applications and websites, otherwise referred to as vibe coding. Mustafa Suleyman, CEO of Microsoft AI said that the new announcements “feel like this is a new era of AI that you control on your terms.” For the past three years, Microsoft’s AI strategy has been closely associated with its multibillion-dollar investment in OpenAI. While that relationship remains intact, in April, Microsoft and OpenAI announced a revised partnership designed to provide greater flexibility for both organisations and Microsoft had already begun to diversify by bringing Anthropic models to 365 Copilot. Microsoft’s announcements yesterday underpin a strategy to develop AI models that it owns, controls and can commercialise independently. CEO Satya Nadella said: “We believe that times come for every company to just move from consuming a frontier model to fully participating at the frontier and the frontier ecosystem.” This news may ultimately be good for end users from a cost perspective. Reuters has previously pointed out that Microsoft is diversifying to help bring costs down. Microsoft’s emphasis on owning more of the underlying technology stack could allow it to integrate governance, security and auditability more deeply into Copilot and Azure-based legal AI solutions. End users will want to be assured, however, that future legal workflows will continue to be able to switch between models and that this announcement provides more optionality, not less.

**Companies:** Microsoft
**Countries:** United States

[Read the full story on Legal IT Insider](https://legaltechnology.com/microsoft-unveils-first-llm-vibe-coding-model-and-speech-text-updates/)

---

Canonical: https://newsio.io/zh-TW/n/d58c6cf6-55db-499d-a3ba-a37705ea4db6/mai-thinking-1-anthropic-claude-opus-4-6
Summarized by Newsio from Legal IT Insider. https://newsio.io/how-it-works