Few months ago I wrote here that Opus 5 is the closest thing to AGI I had on my mac. Yesterday Google answered. Gemini 4 Argon is now open for developers, businesses and normal users, not only the small security partner group it started with at the end of September.
What caught my eye is not the benchmark table (every launch has a table where they win). It's the price. Introductory rate is $2 per million input tokens and $10 output. On the Vals AI index it came #1 of 41 models at 68.9%, and one test there cost about $15.68, against $32.14 for Claude Opus 5.5. Top spot for half the money - sounds familiar?
Saw it in The Neuron recap this morning, this breakdown has all the numbers: https://andrew.ooo/answers/what-is-gemini-4-argon-google-frontier-model-october-2026/
But. Bloomberg had people inside Google saying it still stumbles on real coding, especially front-end. Google says that's inaccurate. For a small team that lives in Next.js and React, front-end is exactly the part that matters.
So for STARTUPS the real news is not who is smartest this week. Its that picking a model is now a cost decision, not a loyalty one.
I'm not switching anything until I try it on real work myself.
Get an email when Gediminas publishes