Google already uses Gemini 3.5 Pro, but it makes you wait

For a while now I've been watching how every model "launch" gets celebrated as if the thing were born that day. CNBC ran a headline saying Google was rolling out new models and personal agents to keep from falling behind OpenAI and Anthropic; TechCrunch reported that with Gemini 3.5 Google is betting its next wave on agents; and Google's own blog introduced the 3.5 family as "frontier intelligence with action." It all sounded like a premiere.

At I/O on May 19, Sundar Pichai presented Gemini 3.5 Pro and asked from the stage to "give us until next month to ship it to you." The detail: the model wasn't waiting in a drawer. Google admits it already uses it internally and in restricted preview. What we call a "launch" isn't the day the model exists, it's the day they decide to hand you a version they've been using for themselves for a while.

I think there's a perverse game going on here. The labs hold their models back and put them out when the competition does: it's not about being first or last, it's about not falling off the board. Sometimes they don't even launch; they just say they're launching to stay in the game. It's a tease. And these jokes seem to cost nothing, but they burn your image: every "give us until next month" shows that the calendar is set by the rival, not the product. Sundar Pichai, you're not taking this seriously.

Sources: CNBC · TechCrunch · Google · MacRumors (Pichai quote)