Gemini 4 Argon.

Every frontier launch promises better reasoning. Gemini 4 Argon, Google DeepMind’s first model of the Gemini 4 generation, makes a more specific bet: that the real bottleneck is stamina. Announced on 30 September 2026 and rolled out broadly the following month, Argon is built to sustain deep, multi-step work — migrating entire codebases, grinding through legal and financial analysis, hunting vulnerabilities — pairing a 1 million token context window with up to 1 million output tokens in one response, up from the previous generation’s 64K output ceiling. This review covers what Argon does, Google’s published benchmark numbers (including the ones it loses), what API access costs, and who should reach for it.

What Is Gemini 4 Argon?

Gemini 4 Argon is Google DeepMind’s flagship reasoning model, available through the Gemini API under the model code gemini-4-argon. It accepts text, image, video, audio and PDF input and returns text, so it can work directly from documents, charts and long recordings. Google positions it at three kinds of hard work: real-world software engineering, enterprise knowledge work such as legal and finance, and defensive cybersecurity. Argon’s first users were trusted cyber defenders in Google’s Fairwind Program, alongside a voluntary U.S. government pre-release review, before access widened. Inside Google, thousands of employees use it for coding and research, including migrating C/C++ codebases of up to 800,000+ lines to Rust.

Key Features

• 1M-token unified context: A context window of 1,048,576 tokens shared across input and output, so large codebases and document sets fit in one request.
• Up to 1M output tokens: With long decode enabled, one response can reach 1,048,576 tokens (65,536 by default) — enough for complete multi-file applications without stitching truncated replies.
• Adjustable thinking levels: Reasoning effort can be set to low, medium or high, trading depth against latency and cost.
• Full tool-use toolkit: Function calling, structured outputs, code execution, file search, URL context and grounding with Google Search and Maps, plus computer use in preview.
• Frontier benchmark results: Google reports 77.9% on DeepSWE v1.1 for long-horizon software engineering, first place on the Vals Index (68.9%), #1 on AutomationBench (51.3%), and 91.7% on the LVBench long-video benchmark.
• Built for cyber defense: Argon can autonomously find, validate and patch vulnerabilities, ties for first on CWE-bench v1 at 68%, and is Google’s most resilient model yet against indirect prompt injection.

How to Use Gemini 4 Argon

1. Get API access in Google AI Studio and enable billing — Argon is a paid-tier model with no free tier.
2. Select the model gemini-4-argon in your API call or in AI Studio’s model picker.
3. Load your context — a repository, contracts, filings or a long video — and set the thinking level to match the task.
4. Enable long decode when you need very large single responses, such as a full refactor or audit document.
5. Or subscribe instead: Google AI Pro and Ultra subscribers can use Argon in the Gemini app, and it is also offered through Google’s Antigravity platform.

Gemini 4 Argon Pricing

Argon is API-first and paid only — Google’s official pricing page lists no free tier for this model. It launched at introductory rates that run through 30 November 2026; from 1 December 2026, prices double. All prices are per 1 million tokens in USD, from the official Gemini API pricing page:

Plan	Price	What you get
Standard (intro)	$2 in / $10 out	Full Argon access via the Gemini API and AI Studio at introductory rates, through 30 November 2026
Standard (from 1 Dec)	$4 in / $20 out	The same standard API access once the introductory period ends and prices double
Batch / Flex	$1 in / $5 out	Half-price asynchronous processing for non-urgent workloads (also doubling from 1 December 2026)
Cached input	$0.10	Per 1M cached input tokens at introductory rates — 95% off standard input; implicit caching only

Search grounding includes 5,000 free requests per month on paid plans, then $14 per 1,000 requests. Consumers can instead use Argon via Google AI Pro and Ultra subscriptions in the Gemini app.

Pros and Cons

Pros

• The 1M output limit is in a class of its own — whole projects and reports in a single response
• Unified 1M-token context handles entire codebases and document collections at once
• Leads Google’s published tables on coding (DeepSWE v1.1), knowledge work (Vals Index) and AutomationBench
• Introductory pricing undercuts rival frontier models, and Batch/Flex halves it again for async work
• Genuinely multimodal input — text, images, video, audio, PDFs — with strong long-video understanding

Cons

• No free tier; casual users must rely on a Pro or Ultra subscription instead
• Prices double on 1 December 2026, so the launch economics are temporary
• Not best at everything: Google’s own table puts Claude Opus 5.5 ahead on Terminal-Bench 4.0 (66.4% vs 57.4%)
• Text output only — no image generation and no Live API support
• Filling the full 1M output window costs $10 even at introductory rates

Who Should Use Gemini 4 Argon?

Argon suits teams whose tasks are long, not quick: engineering groups doing large refactors and legacy migrations, legal and finance teams running multi-step research across big document sets, and security teams who want a model that finds and patches vulnerabilities autonomously. For everyday chat, short coding help or tight budgets, a smaller Gemini model — or a rival flagship such as Claude Opus 5.5 or GPT 6.1 Sol — may be the better daily driver, with Argon reserved for jobs that genuinely run long.

Frequently Asked Questions

How much does Gemini 4 Argon cost?

It costs $2 per million input tokens and $10 per million output tokens at introductory rates through 30 November 2026, doubling to $4 and $20 from 1 December 2026. Batch and Flex cost half the standard rate.

Is Gemini 4 Argon free to use?

There is no free tier for Argon in the Gemini API — you pay per token on a paid plan, or use a Google AI Pro or Ultra subscription in the Gemini app.

What is Gemini 4 Argon’s context window?

The API documentation lists a unified window of 1,048,576 tokens shared between input and output. One response can reach 1,048,576 tokens with long decode enabled; the default synchronous limit is 65,536.

How is it different from earlier Gemini models?

It is the first model of the Gemini 4 generation, aimed at sustained long-horizon work. The headline change is output length — from 64K tokens to up to 1M — alongside stronger coding, knowledge-work and cybersecurity results.

Can Gemini 4 Argon generate images?

No. Argon accepts images, video, audio, PDFs and text as input, but its output is text only; image generation and the Live API are not supported for this model.

Final Verdict

Gemini 4 Argon is Google’s most serious frontier play yet: a model designed to keep working — and keep writing — long after rivals hit their limits. The 1M-token context and output, leading scores on coding and knowledge-work benchmarks, and aggressive introductory pricing make it the model to beat for long-horizon professional work. Budget for the December price rise, and note that even Google’s own tables show rivals winning a few rounds — for deep work at scale, Argon is the one to try first.

Visit Gemini 4 Argon