Google launched Gemini 4 Argon broadly at $2/$10 intro pricing, doubling December 1

Google sells Gemini frontier models through APIs, AI Studio, consumer subscriptions and Gemini Enterprise.

That a model ranked mid-pack on general intelligence can win enterprise workflows and developer share on price — $2/$10 per 1M tokens, doubling from December 1.

Google unveiled Gemini 4 Argon on September 30, 2026, restricting early access to approved cyber defenders and a voluntary US government pre-release process while safety testing continued.

On October 13 it opened Argon broadly through the paid Gemini API, AI Studio, Antigravity and Gemini Enterprise at introductory $2 per million input tokens and $10 output — half Claude Opus 5.5's list price, one-fifth of GPT-6 Astra's — expiring November 30, then doubling to $4/$20.

The introductory $2/$10 pricing has a published expiry — November 30, 2026 — turning the cheap window into a commitment device for developers deciding where to build.

Google competed on professional-work benchmarks where Argon leads (Vals Index 68.90%) rather than aggregate intelligence rankings where it trails Opus 5.5.

The restricted September rollout — Fairwind cyber defenders plus US government pre-release access — sequenced safety review into the launch.

Google's own engineering served as the reference customer: the Chrome refactor in weeks, 40% faster ADK session prep, and Gemma throughput gains are all self-reported proof points.

Limited-time launch pricing is a land grab: the developers who build during the cheap window become the switching costs that survive the December doubling.

As of October 13, 2026, Argon is broadly available at introductory pricing that expires November 30; from December 1 the standard API price is $4/$20 per million tokens. Google is expanding Fairwind access to hundreds more approved cyber partners, including versions with fewer restrictions for authorized defensive work. Independent evaluations split by task type — first on Vals's professional index, mid-pack on Artificial Analysis's overall index — and reviewers caution the surrounding Antigravity tooling still trails Claude Code and Codex.