Google took the velvet rope down on Tuesday. Gemini 4 Argon, the flagship AI model the company had kept locked behind a defenders-only program for two weeks, is now available to developers and businesses worldwide — and it arrives with a price tag that is clearly designed to start a fight. According to Channel News Asia, Argon opens to paying customers at $2 per million input tokens and $10 per million output tokens, with cached inputs discounted 95% — rates set to double after November 30.
The model itself has been public knowledge since The Verge reported its unusual debut on October 1: no developer access, no consumer rollout — just a hand-picked circle of cybersecurity partners inside Google’s Fairwind Program, plus a phased U.S. government pre-release review. The message then was caution. The message now is volume.
The Price Is the Pitch
Forget the benchmark tables for a minute, because Google clearly wants you to look at the receipt instead. At $2/$10, Argon undercuts both flagship rivals on paper: Anthropic’s Claude Opus 5.5 charges $4 and $20 per million input and output tokens, while OpenAI’s GPT-6 Astra runs $10 and $50. Per-token, Argon is a fifth of Astra’s price and half of Opus 5.5’s — and independent scoring puts the gap even starker per completed task, with Argon costing roughly 60% of Astra on the Artificial Analysis Intelligence Index.
The catch is the calendar. The introductory rates expire November 30, when they double to $4/$20 — matching Opus 5.5 exactly. Read that as a six-week land-grab: Google is subsidizing the experimentation phase, betting that once developers build on Argon during the cheap window, they won’t rip it out in December.
What the Open Gates Actually Unlock
Google’s announcement post said the model is available now to Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search, to Ultra subscribers in the Antigravity coding environment, and to developers and enterprises through the Gemini API and AI Studio. The API carries a combined 1-million-token input/output ceiling — up from a 64,000-token output cap — handled by a new continuation protocol for unusually long jobs.
The two weeks of gated access produced the first real-world data points. A Google Cloud engineering account described an overnight coding run in which Argon proposed six changes: two passed review and shipped, four did not. Google also says more than 100,000 of its own employees already use the model monthly, and that internal teams running it in Antigravity increased agentic code submissions by over 35% while cutting code rollbacks by nearly a third. Named early enterprise testers include Deloitte, LexisNexis, Ambience Healthcare, and Isomorphic Labs.
Why this matters
Here’s what the straight-release coverage misses: this launch is Google buying its way back into a conversation it lost. The last few months were rough for the company’s AI credibility — it scrapped plans to release Gemini 3.5 Pro after CEO Sundar Pichai had publicly promised it for June, and it overhauled DeepMind’s leadership after co-founder Demis Hassabis stepped aside and several Gemini model leads departed. Argon’s job was never just to win benchmarks; it was to prove the new regime can ship. The promotional pricing is the tell: Google is confident enough to compete on cost, and nervous enough to need the customers locked in before the window closes.
The second thing to watch is whether the numbers survive contact with real work. Two weeks ago, we reported that Google’s own engineers were skeptical of Argon’s benchmark story, particularly on coding. Until Tuesday, nobody outside Google could check. Now developers finally can — the overnight-coding scorecard (two shipped, four rejected) is the first public hint at what that audit looks like, and it reads more honest than a leaderboard.
Meanwhile in AI Tech: our deep dive on Argon’s benchmark wins and the staff skepticism that shadowed the gated launch.
Meanwhile in Tech News: the White House AI meeting with tech CEOs — the policy backdrop Google has to navigate as it opens the gates.
