Microsoft has launched a suite of AI models at the Build 2026 developer conference, intensifying its competition with Anthropic and OpenAI. Google CEO Sundar Pichai acknowledged that the company is falling behind in agentic coding, while Anthropic announced a significant expansion of its cybersecurity initiatives alongside its IPO plans.
Microsoft Declares Independence from OpenAI, Unveils Seven MAI Models
According to reports from the IT industry, Microsoft unveiled seven AI models it developed in-house during the Build 2026 conference, which took place in San Francisco from June 2. This move is seen as a strategic effort to reduce its long-term dependence on external model providers like OpenAI.
The flagship model, MAI-Thinking-1, features 35 billion active parameters and a context window of 256,000 tokens, emphasizing that it was developed without distilling from third-party models. It was rated on par with Claude Opus 4.6 in the SWE Bench Pro coding benchmark. The coding-specific model, MAI-Code-1, has been integrated into GitHub and VS Code, while MAI-Transcribe-1.5 supports voice recognition in 43 languages. Microsoft has also revised its contract with OpenAI to set revenue-sharing limits and terminate exclusive sales rights.
Google's Pichai Acknowledges Lag in Agentic Coding
In an interview with the New York Times Hard Fork podcast, Google CEO Sundar Pichai publicly admitted, "We are currently somewhat behind in agentic coding, tool usage, and long-term task execution."
At Google I/O, the company introduced a developer subscription tier priced at $100 per month as part of its low-cost strategy. Pichai explained that while Google excels in generating single-use web interfaces, competitors have the advantage in handling large codebases for long-term projects. He cited Anthropic's collaboration with Cursor and analyzed that the lack of developer-facing products has created a gap in data feedback loops.
Anthropic Expands Project Glasswing to 150 Organizations
Anthropic announced it will expand Project Glasswing to over 150 organizations across more than 15 countries. New sectors included in the initiative are energy, utilities, healthcare, telecommunications, and hardware. Partners reportedly include Samsung, SK Hynix, SK Telecom, NATO, the EU cybersecurity agency ENISA, and Okta. Initial findings from about 50 partners identified over 23,000 potential vulnerabilities, with more than 6,000 classified as severe or critical.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.