GOOGLE GEMMA 4 AND MICROSOFT’S THREE-MODEL CHALLENGE

The open-weights landscape shifted dramatically this week. Google DeepMind released Gemma 4 on April 2, calling it “byte for byte, the most capable open model” — built from the same research as Gemini 3. Four sizes are available, all under Apache 2.0 license. This means developers can use, modify, and deploy commercially without restrictions.

Not to be outdone, Microsoft announced three in-house foundational models on the same day. The Verge described this as Microsoft “shivving OpenAI” — its own AI partner — by building competing speech and image models. TechCrunch called it a “direct shot at OpenAI and Google.”

VentureBeat’s analysis noted that Microsoft’s move signals a fundamental shift: the company no longer wants to rely solely on OpenAI for core AI capabilities. The three models cover:

• Speech transcription (state-of-the-art accuracy)
• Image generation and understanding
• Multimodal reasoning

For developers, the implications are significant:

• Open-weights models like Gemma 4 reduce dependency on paid APIs
• Competition drives down costs across the board
• Microsoft’s models may integrate deeper into Azure services
• The definition of “state-of-the-art” changes weekly

The technology race is now a three-way battle: Google’s open approach, Microsoft’s enterprise integration, and OpenAI’s frontier models. Each path offers different tradeoffs for developers.

Sources:

• Google DeepMind: Gemma 4 announcement
• TechCrunch: Microsoft’s three models
• The Register: Microsoft shivs OpenAI
• VentureBeat: Model competition analysis