fetchpriority=”high” alt=”Microsoft Experience Center, New York. Image: Shutterstock/Decrypt” width=”1600″ height=”900″ decoding=”async” data-nimg=”1″ class=”” style=”color:transparent” srcSet=”https://img.decrypt.co/insecure/rs:fit:1920:0:0:0/plain/https://cdn.decrypt.co/wp-content/uploads/2026/02/microsoft-gID_7.jpg@webp 1x, https://img.decrypt.co/insecure/rs:fit:3840:0:0:0/plain/https://cdn.decrypt.co/wp-content/uploads/2026/02/microsoft-gID_7.jpg@webp 2x” src=”https://img.decrypt.co/insecure/rs:fit:3840:0:0:0/plain/https://cdn.decrypt.co/wp-content/uploads/2026/02/microsoft-gID_7.jpg@webp”>
Microsoft has officially entered the race to develop proprietary frontier AI models, unveiling seven new in-house developed systems during its annual Build conference. The tech giant claims its flagship reasoning and image generation models surpass key competitors from Anthropic and Google in performance benchmarks. This strategic move signifies Microsoft’s ambition to become a leader in AI development, beyond its role as a major investor and infrastructure provider for OpenAI.
Key Takeaways
- Microsoft announced seven new AI models, including reasoning, coding, image, transcription, and voice generation systems.
- The company claims its MAI-Thinking-1 reasoning model outperformed Anthropic’s Claude Sonnet 4.6 and matched Claude Opus 4.6 on a leading coding benchmark.
- Microsoft’s MAI-Image-2.5 models reportedly exceeded Google’s Nano Banana 2 in image-editing benchmarks.
- This launch represents a significant step in Microsoft’s strategy to build its own advanced AI capabilities, complementing its partnership with OpenAI.
- The AI landscape continues to heat up, with major players like Anthropic and Google also releasing new, advanced AI models and expanding their offerings.
At the forefront of this announcement is MAI-Thinking-1, Microsoft’s proprietary foundation model for reasoning. According to Microsoft AI CEO Mustafa Suleyman, blind evaluations by independent testers favored MAI-Thinking-1 over Anthropic’s Claude Sonnet 4.6. Furthermore, the model achieved a 97% score on the AIME 2025 benchmark, indicating advanced problem-solving and reasoning capabilities, and performed comparably to Claude Opus 4.6 on a challenging coding benchmark.
Seven new models launching at Build: let’s go!
Reasoning. Code. Image. Transcribe. Voice.Built from scratch on a clean data lineage, designed for efficiency, working seamlessly as a family of models
Thread 🧵 #MSBuild
The newly introduced suite also includes MAI-Code-1-Flash, a streamlined coding model designed for integration with GitHub Copilot and Visual Studio Code. For visual tasks, MAI-Image-2.5 and its Flash variant are presented as superior to Google’s Nano Banana Pro for image editing. The lineup is further enhanced by MAI-Transcribe-1.5, capable of processing 43 languages, and MAI-Voice-2, a speech generation model supporting 15 languages and able to mimic a speaker’s voice from a brief audio sample.
Suleyman highlighted the exponential growth in compute power required for training frontier models, noting a trillion-fold increase and predicting another thousand-fold rise in the next three years, which will unlock even more sophisticated AI capabilities. This development occurs amidst fierce competition in the AI sector, with Anthropic recently releasing its Opus 4.8 model and expanding access to its cybersecurity-focused Mythos model. Google, too, has been active, showcasing its multimodal Gemini Omni and task-automating Gemini Spark AI agent at Google I/O.
Microsoft’s strategic push for proprietary AI development, evidenced by the new MAI models, suggests a desire to diversify its AI offerings. The company indicated that its MAI models delivered a superior win rate and were significantly more cost-effective than GPT-5.5, aligning with a stated goal to provide developers and businesses with AI solutions tailored to their specific needs and control requirements.
Long-Term Technological Impact on the Blockchain and Web3 Ecosystem
Microsoft’s significant investment and advancement in proprietary AI models have profound implications for the blockchain and Web3 sectors. As AI capabilities become more sophisticated and accessible, their integration into decentralized technologies is poised to accelerate innovation. For Layer 2 scaling solutions, advanced AI could optimize transaction processing, gas fee prediction, and network security, making them more efficient and user-friendly. In Web3 development, AI can streamline smart contract creation, audit processes, and even power more intelligent decentralized applications (dApps) and decentralized autonomous organizations (DAOs).
The drive for AI that is “on the frontier” and offers greater control could translate into more robust and trustworthy AI-powered oracle services, which are crucial for feeding real-world data into smart contracts. Furthermore, AI’s ability to analyze complex data sets could unlock new insights into on-chain activity, improving fraud detection, risk management, and personalized user experiences within decentralized finance (DeFi) and gaming platforms. This technological synergy between advanced AI and blockchain infrastructure suggests a future where decentralized systems are not only more scalable and secure but also more intelligent and intuitive, pushing the boundaries of what is possible in a decentralized digital world.
Based on materials from : decrypt.co