Pichai: Gemini 4 Will Require Much Larger Base Models

At Alphabet's Q2 2026 earnings call, CEO Sundar Pichai revealed that Google has launched "the most ambitious pretraining run in its history" for Gemini 4. Pichai stated plainly that the next leap in model capabilities will require "significantly larger base models" — larger in both parameter count and training data volume.

Pichai also called out areas where Gemini still lags: "Coding and agentic coding is a great example, and the teams are very, very focused on that." He also highlighted surging demand for the compact Gemini Flash series, which occupies the sweet spot between performance and cost.

1By the Numbers: $195–205B and 950M Users

Alphabet raised its 2026 capital expenditure guidance to $195–205B, up from the prior estimate of $180–190B. Management said demand for AI infrastructure continues to outpace the company's ability to build it out. The Gemini app reached 950M monthly active users as of Q2.

Pichai's remarks fit a broader trend: the frontier AI scaling race shows no signs of slowing. Even as compact, efficient models emerge, leading labs continue to pour enormous sums into training large base models — convinced that those are what will define the next generation of capabilities.