Google Launches Gemini 4 Argon: Frontier AI at a Fifth of Rival Prices

Key takeaway: Google opened access to Gemini 4 Argon on October 13, 2026 — its first major frontier AI model since Gemini 3 in late 2025. Built for long, complex professional work like coding, finance, legal, and cybersecurity tasks, it launches at an introductory price of $2 per million input tokens and $10 per million output tokens — a fraction of what OpenAI and Anthropic charge for their top models.

What Google announced

Google first announced Gemini 4 Argon on September 30, 2026, initially limiting it to vetted cybersecurity teams. Two weeks later, on October 13, the company opened it to developers, businesses, and paying subscribers — its broadest frontier-model release in nearly a year.

Argon is tuned for difficult, long-horizon work: multi-step coding projects, scientific analysis, financial and legal research, and cybersecurity. It supports a combined context of roughly 1 million tokens, and maximum output jumps to up to 1 million tokens — up from 64,000 in earlier Gemini models — which removes a lot of friction when generating long documents or large codebases in one pass.

Google named early enterprise testers including Deloitte, LexisNexis, Ambience Healthcare, and Isomorphic Labs, and expanded its Fairwind Program so hundreds more vetted cybersecurity defenders can use a version of the model with cyber safeguards removed — letting defenders test attacks the way real adversaries would.

Where it wins — and where it doesn’t — on benchmarks

On Google’s own published evaluation table, Argon leads on several professional-work benchmarks: 68.9% on the Vals Index (finance, coding, legal, tax tasks) versus 63.1% for GPT-6 Astra and 67.0% for Claude Opus 5.5; 77.9% on DeepSWE v1.1, a long-horizon software engineering benchmark; and 51.3% on AutomationBench, well ahead of both rivals. It tied GPT-6 Astra on the CWE-Bench cybersecurity benchmark at 68.0%.

Elsewhere the picture is mixed. Claude Opus 5.5 still leads on Terminal-Bench 4.0 (66.4% vs. Argon’s 57.4%), and GPT-6 Astra leads on FrontierSWE v2 (65.5% vs. 55.0%). On the independent Artificial Analysis Intelligence Index, Argon scored 53 on its highest setting — tying GPT-6 Astra but trailing both Claude 5.5 models by about five points, while still a 23-point jump over Google’s previous best, Gemini 3.1 Pro Preview.

Google also shared internal results: engineering teams using Argon inside its Antigravity agentic coding environment increased agentic code submissions by over 35% and cut code rollbacks by almost a third. On Gray Swan’s indirect prompt-injection benchmark, attack success against the model fell to 0.7% across 15 adaptive attempts.

Pricing and how to get access

The introductory pricing is the headline: $2 per million input tokens and $10 per million output tokens, with cached input discounted 95%. That is roughly a fifth of OpenAI’s published price for GPT-6 Astra ($10/$50) and half of Anthropic’s Claude Opus 5.5 ($4/$20) — though OpenAI’s cheaper GPT-6.1 Sol already matches Argon at $2/$10.

The catch: the introductory rate ends November 30, 2026, after which Argon moves to $4/$20 — matching Claude Opus 5.5 and still well below GPT-6 Astra. Access is limited to paying users: the paid tier of the Gemini API (AI Studio, Android Studio), the Gemini platform and Antigravity for enterprises, and the Gemini app and AI Mode in Search for Google AI Pro and Ultra subscribers. Free-tier users don’t get it.

Why it matters: the practical angle

For developers and teams, the math is simple. Long agentic coding runs — exactly what Argon is tuned for — burn a lot of tokens, and cutting per-token cost by up to 80% changes what experiments are affordable. Our earlier coverage of AI agents for business automation shows what these workflows look like in practice, and the custom AI chip race from Microsoft is the other half of why costs are falling.

The broader signal: the frontier-model market is shifting from a pure capabilities race to a price race. With two top-tier models now at $2/$10 per million tokens, the cost of putting serious AI coding assistance into production is falling fast — and while Argon’s introductory price is live through November, it’s a good window to test.

Analysis: what Google is really betting on

USABite analysis — interpretation, not verified fact: Google isn’t trying to win every benchmark with Argon; it’s trying to win developers. By pricing a near-frontier model at a fraction of rivals and tuning it for the exact workloads (long agentic coding runs) that burn the most tokens, Google makes building on its API the cheap default, where lock-in is stickiest. The risk: with benchmark leadership split across all three labs, Argon becomes “cheap and good enough” rather than the automatic choice.

Frequently asked questions

When did Google launch Gemini 4 Argon?

Google announced it on September 30, 2026 with limited cybersecurity-team access, then opened it broadly to developers, businesses, and paying subscribers on October 13, 2026. It is Google’s first major frontier-model release since Gemini 3 in late 2025.

How much does Gemini 4 Argon cost?

The introductory price is $2 per million input tokens and $10 per million output tokens (cached input 95% off) — roughly a fifth of GPT-6 Astra and half of Claude Opus 5.5. After November 30, 2026, it rises to $4/$20, matching Claude Opus 5.5.

Is Gemini 4 Argon better than GPT-6 Astra and Claude Opus 5.5?

It depends on the task. Google’s published results show Argon leading on the Vals Index, DeepSWE v1.1, and AutomationBench; Claude Opus 5.5 leads on Terminal-Bench 4.0 and GPT-6 Astra on FrontierSWE v2. On the independent Artificial Analysis Intelligence Index, Argon ties GPT-6 Astra but trails both Claude 5.5 models by about five points.

How can I use Gemini 4 Argon?

Through the paid tier of the Gemini API in AI Studio and Android Studio, the Gemini platform and Antigravity for enterprises, or the Gemini app and AI Mode in Search as a Google AI Pro or Ultra subscriber. It is not on free Gemini accounts.

What is the context window of Gemini 4 Argon?

Roughly 1 million tokens of combined input and output, with maximum output up to 1 million tokens — up from 64,000 in earlier Gemini models, making it suitable for long documents, large codebases, and extended multi-step jobs.