Google typically reserves its biggest product launch each year for the I/O conference. CEO Sundar Pichai’s restraint this time around says a lot about the company’s position in the AI coding race.
During his keynote, Pichai told the audience that Google’s new flagship Gemini 3.5 Pro AI model is not ready yet, prompting audible groans.
Since I was there too, I spent the rest of the event coming up with a theory for the apparent delay. Google is holding off on this new model for a while to further improve AI coding tasks.
Anthropic’s Claude Code took the world by storm last year, and OpenAI’s Codex has recently gotten even better. These frontier labs are gaining developer mindshare and generating significant revenue by enabling programmers to automate and accelerate coding tasks using agents and AI tools.
This is a revolution that is about to transform Silicon Valley, but Google was probably a little late. But it doesn’t last long.
Instead of releasing the 3.5 Pro, Pichai talked passionately about another new model, the Gemini 3.5 Flash. This is a smaller model that is only slightly less powerful than the world’s current top models, yet faster and much cheaper.
Google has already made 3.5 Flash the primary model powering its Antigravity AI coding service. Starting today, software developers will be using this tool to churn out code.
This generates tons of anonymous and valuable data. For example, if an engineer starts a new coding project in Antigravity and suddenly stops the task. This suggests that something in the output from Flash 3.5 was incorrect.
Google can use this feedback data to improve the larger 3.5 Pro model, possibly through reinforcement learning. Reinforcement learning is a technique for refining new AI models by rewarding good results and punishing bad ones.
Signals when running Antigravity on smaller 3.5 flash models can aid this process in important ways. That’s because coding is particularly good at generating clear signals for AI model development. If your code is good, it will probably work. In severe cases, they often break things.
This should give the larger 3.5 Pro model a strong clue as to which coding outputs worked and which didn’t.
“I know you can’t wait to get your hands on it,” Pichai said on stage. “Please wait until next month.”
So when this big new model finally arrives, we expect it to significantly improve the power of today’s hottest generative AI application: coding.
Sign up for BI’s Tech Memo newsletter here. Please contact us by email. abarr@businessinsider.com.