Google's New Model Outputs Stand Out, and Its Eval Effort Shows

omarsar0 · x · 2026-11-22

The author notes how much benchmark effort Google put into this model. Benchmaxxing isn't good per se, but Google has likely figured out good ways to evaluate these models and hill-climb on the signals that matter most to users — that's what evals can do.

He earlier observed that this model's outputs (both for interactions and notes) are clearly better than other models', making it easier to interact with, learn from and research ideas.

Related event: Scholars Say Gemini Neon Quality May Replace Opus 5.5 as Daily Driver(4 posts)→

Original post →

More from Models

Models channel →