Gemini 4 Argon offers 1M output tokens and posts several benchmark results
Argon supports 1M output tokens versus 128K for Astra and Opus 5.5, scores 77.9% on DeepSWE long-horizon software engineering against 74.1% and 74.2%, and leads by 8-10 points on business workflows, financial research, legal tasks and long-video understanding.
I’m honestly blown away by Gemini 4 Argon .. 1 MILLION output tokens compared to the standard 128K limit on Astra and Opus 5.5 !! That’s insane .. Just imagine what you could get done in a single output ! I’d been thinking Gemini had fallen behind in the race, but apparently Google was cooking something behind the scenes .. According to Google’s results, it scores 77.9% on DeepSWE for long-horizon software engineering, compared to Astra’s 74.1% and Opus 5.5’s 74.2% .. And it’s 8 to 10 points ahead of both on business workflows, financial research, legal tasks, understanding long videos, and tracking relationships across huge amounts of context .. Then there’s what it’s already doing inside Google .. Its agents optimized Google’s infrastructure and freed up over 300 TiB of memory ! Separately, it improved the resource requirements of a quantum subroutine by 40% against a published baseline In minutes Google’s been cooking .. This is the kind of drop I’ve been waiting for.
This post quotes another post: x.com/google/status/2105388143902175529