
Google releases Gemini 4 Argon with broader access and introductory token pricing
The AMW Read
Broader access extends an already covered release, while introductory pricing and distribution across Google's products make it consequential for frontier-model competition.
Google releases Gemini 4 Argon with broader access and introductory token pricing
Google has released Gemini 4 Argon to Pro and Ultra subscribers through the Gemini app and AI Mode in Search, with access also reaching Ultra subscribers on Antigravity. Introductory pricing is $2 per million input tokens and $10 per million output tokens, doubling after the promotional period ends in November. Google claims stronger reasoning and benchmark results above GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across several disciplines; the report provides no independent validation of those comparisons. Availability for Plus subscribers and free users remains unclear.
The release puts Google back into direct competition for frontier-model workloads through both pricing and distribution. Access across Search, Gemini, and Antigravity gives the model several routes to users, while the introductory token rates challenge rival pricing. This extends AMW's October 1 coverage of restricted cybersecurity access into a broader commercial rollout. Google's reported internal applications, including vulnerability remediation and debugging Wing's drone simulator, illustrate its intended workflow scope, but do not establish comparable reliability for outside customers.
Builders should compare completed-task cost and reliability before switching workloads. The report places introductory token prices at roughly one-fifth of rival rates, but also warns that different token consumption can erase part of that advantage. A concrete evaluation should rerun representative tasks, measure total tokens and successful completions, and calculate costs at both introductory and post-promotion rates. For investors, the question is whether lower pricing and Google's distribution translate into sustained usage; benchmark claims alone cannot answer it.

