Gemini 3.5 Pro: Google threw the model in the trash and started from scratch

Google abandoned the base architecture of Gemini 3.5 Pro and restarted pretraining from scratch. What went wrong, what does this reveal about the AI model race, and what does it mean for developers who were already waiting for this release?

We analyze the reported problems with mathematical reasoning and image generation, DeepMind’s new strategic focus on price and efficiency in competition with OpenAI and Anthropic, and why every week of delay costs Google the loyalty of developers who serve as its enterprise adoption channel.

If you’re interested in the artificial intelligence industry and want to understand the decisions behind the headlines, subscribe and turn on notifications so you don’t miss the follow-up on July 18.

🤖 AI-generated content: the script, voices, and images in this episode were produced using artificial intelligence tools.

📷 Images:

• “Google First Production Server” — Takuya Oikawa (CC BY-SA 2.0) — https://www.flickr.com/photos/34597916@N00/313528682
• “Supercomputing” — Idaho National Laboratory (CC BY 2.0) — https://www.flickr.com/photos/30369883@N03/6236052094
• “hackers at work” — Dave_B_ (CC BY 2.0) — https://www.flickr.com/photos/54829270@N00/5965479074
• “eCommerce Developers” — Infusionsoft (CC BY-SA 2.0) — https://www.flickr.com/photos/30759596@N02/4443187805
• “My desktop at the moment” — bionicteaching (CC BY-SA 2.0) — https://www.flickr.com/photos/29096601@N00/16598625858

#Gemini35Pro #GoogleAI #ArtificialIntelligence #DeepMind #OpenAIvsGoogle #AIModels #AI2025