Story
Oct 13, 2026
Google’s Gemini 4 Argon Returns to the AI Fight—But Behind a Paywall
Google sees Gemini 4 Argon as a credible route back to the frontier of AI; its own rollout strategy, however, makes that comeback a premium proposition rather than a broad public release.
Google first previewed Gemini 4 Argon last month, limiting access to selected “trusted cyber defenders” while it tested for security risks. The cautious start came after a period in which OpenAI and Anthropic had appeared to pull ahead—and after Google passed on launching its previously promised Gemini 3.5 Pro.1
On Monday, Meta added to the pressure by announcing its Muse Wonder model family. A day later, Google made its countermove: Argon began rolling out globally to paid API customers, enterprises, Pro and Ultra Gemini subscribers, and Ultra users of its Antigravity development platform. “Gemini 4 Argon is here,” CEO Sundar Pichai wrote, spelling out an access list that notably excludes free users and leaves the timetable for lower-priced Plus subscribers unclear.
2
Google’s case is that Argon is more than a catch-up release. The company says the model delivers “frontier reasoning and improved performance across complex workflows,” with benchmark gains over named OpenAI and Anthropic rivals in coding, cybersecurity, science, finance and law. It has also deployed the system internally for vulnerability patching, YouTube comment-processing efficiency and Wing drone-simulator debugging.3
The commercial pitch is as aggressive as the technical one: an introductory $2 per million input tokens and $10 per million output tokens, roughly a fifth of the cited rivals’ prices. That rate will double after November, though lower token prices do not automatically mean lower real-world costs.3
Google DeepMind product lead Tulsee Doshi called Argon a “powerhouse” and said it is the first in a coming family. Internally, Google says it helped lift AI-generated code production by 35%; the company is also emphasizing safeguards that monitor a model’s chain of thought for concerning behavior.1
Pichai’s demonstrations focus on the practical proof Google now needs: stronger coding and multi-step tool use, from turning a raw Pixel CAD file into a polished animation to handling complex workflows in Antigravity.4 The frontier race is no longer a two-company contest—but Google’s bid to rejoin it begins with a checkout screen.