Gemini 4 Argon Goes Global: Google's Flagship Model Opens to All Paying Users Today

Two weeks ago it was reserved for a small group of cyber defenders. As of today, any Google AI Pro or Ultra subscriber — and every enterprise — can use the most powerful model Google has ever shipped.

On October 13, Google lifted the access restrictions on Gemini 4 Argon: the model is now live in the Gemini app for Google AI Pro and Ultra subscribers, available to enterprises worldwide, and callable by developers in Google AI Studio, Android Studio, and the agentic developer platform Antigravity. On Vertex AI it appears as gemini-4-argon. CEO Sundar Pichai confirmed the global rollout on X the same day.

From Fairwind-Only to Full Availability

Argon's release path was unusual. On September 30, Google DeepMind SVP Koray Kavukcuoglu first unveiled the opening model of the Gemini 4 generation — but initial access went exclusively to the Fairwind Program, a cyber-defense initiative for governments and trusted partners. Handing a frontier model to security defenders first says a lot about where Argon's strengths lie.

The industry had expected a Gemini 3.5 Pro in the first half of the year; Google skipped it and bet directly on Gemini 4. Argon is the first move of that bet.

Specs and Pricing: A Deliberate Price Gap Against Rivals

Argon arrives with two headline specs: a 1-million-token context window for long, multi-step work, and 262k max output tokens.

The pricing strategy is aimed squarely at the competition:

Introductory rate (now through November 30, 2026): $2 per million input tokens, $10 per million output tokens, with cached input at 95% off
Standard pricing (after November 30): $4 per million input, $20 per million output

For comparison, Claude Opus 5.5 sits at $4/$20, and OpenAI's GPT-6 Astra commands $10/$50 — Argon's intro price is one-fifth of Astra's.

Benchmarks: Ahead on 13, Behind on 4

Across the 19 benchmarks Google published against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5, Argon leads on 13. Highlights:

DeepSWE v1.1: 77.9% — Google calls it a new record in real-world software engineering
Vals Index: 68.90%, ranked #1 of 41 models (independently measured by Vals AI, at $15.68 per test — below Sonnet 5.5's $21.34 and Opus 5.5's $32.14)
#1 on Finance Agent v2 (65.40%), a perfect 100% on IOI (tied with GPT-6 Astra), #1 on Zapier's AutomationBench (51.3%)
Harvey's legal agent evaluations: substantial gains over Gemini 3.1 Pro in corporate M&A, data privacy, cybersecurity, and EC/VC work

It is not a clean sweep: Argon trails on FrontierSWE v2 and Terminal-Bench 4.0, posts 45.3% on the PostTrainBench ML engineering benchmark (Opus 5.5: 49.3%), and scores 69.2% on the OSWorld-2.0 offline subset against Astra's 72.6%.

A dose of caution is warranted: most figures are vendor-reported, and no third party had fully reproduced them at launch. The evaluation community has real concerns about benchmark contamination — Epoch previously flagged 23 flawed tasks in DeepSWE.

Security Is the Real Battleground

Argon's most unusual selling point is cybersecurity. Google says it can autonomously find, validate, and patch critical software vulnerabilities across complex codebases spanning 20 programming languages. On Wiz's black-box penetration testing benchmark, Argon outperforms 3.8 Flash Cyber in attack-surface discovery, vulnerability identification, and proof-of-concept generation; it ties for first on the CWE-bench v1 remediation benchmark at 68%.

One notable detail: Google deliberately kept monitoring findings out of training data so Argon's reasoning would not be shaped to evade the monitors — a level of transparency few labs publish. Customer-side validation is following: HubX reports Argon identified 100% of top-priority vulnerabilities in its codebase security audits with 0% fabricated findings.

Already Working Inside Google

Google says Argon has been in internal use for weeks: optimizing Python codebases, patching security holes, making its open Gemma models three times faster at tasks, and helping optimize data centers — freeing up hundreds of terabytes of memory without additional capex. In physical AI, Argon helped ground humanoid robotics research in real-world physics and surroundings.

One Move, a Three-Way Race

With Argon going global, the frontier race is back to a three-way standoff: GPT-6 Astra ($10/$50), Claude Opus 5.5 ($4/$20), and Gemini 4 Argon ($4/$20 — half price during the intro window). Beyond price, the real differentiation is focus: OpenAI bets on consumer agents, Anthropic defends enterprise coding, and Google points its flagship spear at long-horizon reasoning plus autonomous cyber defense.

For developers, this is the cheapest window to evaluate Argon: the introductory rate ends November 30, when prices double.

Sources: Google official blog, Google Cloud Blog, Reuters, CNET, Mashable, Engadget, Vals AI (October 13, 2026)