Google spent 2025 looking like the AI lab that finally woke up, and now it's getting dunked on by the competition for delays that would have been a footnote six months ago.
The Summary
- Google delayed its Gemini 3.5 Pro frontier model while OpenAI and Anthropic shipped new top-tier models, prompting public mockery from Meta's chief AI officer and an OpenAI researcher
- Raymond James analyst calls it a perception shift "from leading edge to trailing edge" despite Google's strong business fundamentals
- Google announced pre-training on Gemini 4 has begun and shipped three faster, cheaper models this week, betting on efficiency over frontier performance
The Signal
The AI lab hierarchy is getting reshuffled in real time, and Google is learning that momentum in this market has a half-life measured in weeks, not quarters. After Gemini 3 put them back in contention at the end of 2025, the delayed 3.5 Pro release has opened a window for competitors to reclaim narrative control.
Meta's chief AI officer writing "gemini who?" on X isn't just trash talk. It's a signal that the competitive dynamics have shifted from technical benchmarks to perception management. When your rivals feel comfortable mocking you publicly, you've lost something more valuable than a leaderboard position. You've lost the fear factor.
"The pushback has shifted perception from leading edge to trailing edge."
What makes this particularly sharp is the timing. OpenAI and Anthropic both shipped new frontier models in recent weeks while Google pushed back its release. In the agent economy, frontier model capability matters because it sets the ceiling for what's possible. If your model can't reliably handle complex multi-step reasoning, you can't build the agents that will actually automate knowledge work. Efficiency models are great for production deployment, but they don't move the Overton window on what agents can do.
Google's response strategy is telling. They're betting on a different game: faster, cheaper models that can run at scale. Three new efficiency-focused models this week, pre-training announced on Gemini 4, and the implicit message that frontier racing isn't the only path to winning the agent economy. It's a reasonable bet if you have Google's distribution and infrastructure. YouTube, Search, and Cloud are all printing money and benefiting from AI integration. The business fundamentals are strong.
But here's the tension: developer mindshare moves faster than enterprise contracts. If the cutting-edge agent builders start defaulting to OpenAI or Anthropic because those models can handle the hardest problems, Google risks becoming the safe, boring choice. The model you use in production after you've proven the concept elsewhere. That's a profitable position, but it's not a throne.
The Implication
Watch what happens in the next 60 days. If Google ships Gemini 3.5 Pro and it's genuinely ahead of the current OpenAI and Anthropic models, this delay becomes a footnote. If it ships and it's merely competitive, the perception problem hardens into reality. The company that invented the transformer architecture doesn't get to be "good enough."
For builders in the agent space, this is a reminder that your infrastructure choices have a shelf life. Model leadership changes hands quarterly now. Build for portability. The best model for your agent today might not be the best model in September.