Summary
Google’s Gemini 4 Argon is now available to users on higher-end paid plans, and Matt Wolfe breaks down what is new and how it performs. Announced about two weeks earlier with access limited to a few cyber defenders, Argon is designed for the most complex, long-running tasks rather than quick question-and-answer use.
The video reviews the early benchmark results. Argon is described as state-of-the-art on the Vals Index for professional tasks across finance, coding, legal work, and taxes, as well as on Automation Bench, legal work, and Deep Suite for agentic coding. It scores 53 on the Artificial Analysis index, on par with Fable but behind Opus 5.5, Sonnet 5.5, and Muse’s new Wonder model, while costing noticeably less per task. Argon also offers an output limit of 1 million tokens, roughly 750,000 words, which Wolfe says no other model currently matches.
Wolfe then puts it to work in Google’s Antigravity coding platform, which offers low, medium, and high modes. Tests include SVG generation, a one-shot Mega Bonk-style 3D game clone, and an interactive physics playground with adjustable gravity presets. He finds the game mechanics strong, though he still rates Opus 5.5 and GPT-6 Astra higher on visual design.
📺 Source: Matt Wolfe · Published October 13, 2026
🏷️ Format: News Analysis
