Understand Google’s new frontier model and start using it well in the Gemini app and API
Short answer
A free Google Gemini 4 Argon guide for October 2026: what Argon is, the phased rollout, Gemini app access on Google AI Pro and Ultra, tasks mode, app connections, Antigravity coding, API model ID and pricing, the 1M-token context and output, benchmarks, safety, prompts, and limits.
Gemini 4 Argon is Google’s new frontier AI model, announced on September 30, 2026. Google calls it its frontier model for real-world coding, enterprise knowledge work, and cyber defense. It is a new model, not a fine-tune of an earlier one.
It is built for:
Long, multi-step tasks
Complex software engineering
Legal, finance, and other professional work
Charts, long videos, and many documents at once
Creative writing
Argon is built to finish long jobs, not just answer quick questions.
02
How it rolled out
September 30
Trusted cyber defenders and internal Google teams first
Next
Paid API customers and Google AI Ultra subscribers
October 13
Gemini app access for Google AI Pro and Ultra subscribers
Google released it in phases for safety. Free Gemini users were not included at launch.
03
Key specs
Context window
About 1 million tokens
Output
Up to 1 million tokens, up from 64,000 before
Inputs
Text, images, video, audio, and PDFs
Output type
Text
04
What is new in the Gemini app
More direct answers, with fewer caveats and stock phrases
More progress updates during long work
A tasks mode toggle, so Gemini can keep working while you are away
Connections to more than 40 apps, including Instacart, OpenTable, Adobe, and Shopify
Easier switching to Google Antigravity’s agentic coding, now in the Gemini macOS app
05
For developers
Model ID
gemini-4-argon
Where
The paid tier of the Gemini API in Google AI Studio and Android Studio, and Google Cloud for enterprises
Launch price
$2 per million input tokens and $10 per million output tokens
Later price
$4 input and $20 output after the introductory period
Caching
Cached input tokens cost 95% less
06
How it performs
Google reports that Argon:
Set a new record on DeepSWE v1.1 for software engineering, at 77.9%
Ties for first on the CWE-bench cybersecurity benchmark
Leads a benchmark for resisting indirect prompt injection
These are Google’s own results. Test it on your own work.
07
Best uses
Big coding projects and migrations
Research across many long documents
Financial and legal analysis, with expert review
Turning long videos and charts into insights
Multi-step tasks across your connected apps
08
Write better tasks
State the goal and what done looks like
Share the files and links it should use
Set rules, budgets, and deadlines
Ask for progress updates
Tell it what needs your approval
Review these five contracts. List every renewal date, notice period, and price increase in one table, flag unusual terms, and cite the page for each. Do not contact anyone.
09
Safety and privacy
Approve anything that buys, books, sends, or shares
Connect only the apps a task needs
Check your Gemini Apps activity and data settings
Do not share passwords or sensitive personal data
Have experts review legal, medical, and financial outputs
10
Know its limits
Access depends on your plan and country
Long tasks and huge outputs use more of your limits or API budget
It can still make mistakes, so verify important facts
Benchmarks are mostly self-reported so far
11
Master Argon prompt
Goal: [what you want done].
Context: [who you are, files, and links].
Rules: [limits, budget, deadline, tone].
Output: [format and length].
Work step by step, give me progress updates, cite sources, and ask before any action that sends, buys, books, or shares.
The golden rule
Do not use Argon for quick questions a smaller model can answer. Give it the long, hard jobs, and check the results.
Frontier
Tasks
Apps
Antigravity
API
1M tokens
Verify
Give Argon one long task you have been putting off this week, with a clear goal and rules.
It has 11 sections, starting with “What is Gemini 4 Argon?”, “How it rolled out”, “Key specs” and “What is new in the Gemini app”, plus 7 more sections.
How long does it take to read?
About 8 minutes. Every section is a short, self-contained card, so you can jump straight to the part you need.
Is this guide free?
Yes. The full guide is free to read on this page, with no sign-up.