A practical field note on Google’s Gemini 4 Argon: what it is, who can use it, how to pick it in the Gemini app, and the exact one-prompt tests from my reel with honest results.
If you came here from my reel and commented ARGON, this is the guide I promised: what Google’s Gemini 4 Argon is, who can use it, how to pick it in the Gemini app, and the exact prompts behind the two tests in the reel.
Read Google’s announcement and the Gemini app availability post. Availability changes, so treat Google’s pages as the final word.
What Gemini 4 Argon is
Google announced Gemini 4 Argon on 30 September 2026 as its new frontier model. Google says it is built for complex, multi-step work across real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence.
In the Gemini app, Google says Argon answers questions more directly, with less hedging, and shares its progress more often on longer tasks.
Who can use it
- Gemini app: Google says Google AI Pro and Ultra subscribers can use Gemini 4 Argon in the Gemini app from 13 October 2026.
- Developers: the Gemini API model code is
gemini-4-argon. It is on the paid tier only, not the free tier, in Google AI Studio and Google’s Gemini platform.
How to pick it in the Gemini app
- Open the Gemini app on a Google AI Pro or Ultra account.
- Open the model picker and choose 4 Argon.
- For tests like mine, turn on Canvas so the result opens as a working page you can click.
- Paste one prompt and let it run. I left the thinking level on its default, Low.
Test 1: an interactive 3D keyboard switch
This is the exact prompt I used, in one message, with no follow-ups:
Build an interactive 3D model of a mechanical keyboard switch (Cherry MX style) as a single self-contained web page using three.js. Model the keycap, upper housing, stem, spring, lower housing and the two metal contact leaves as separate parts. Let me orbit with the mouse, click the keycap to press it down (spring compresses, contacts close, a small "ACTUATED" indicator lights up), and use a slider to explode the parts apart vertically. Label every part. Dark background, clean studio lighting.
What happened: about three minutes later Canvas showed a page called “Interactive 3D Cherry MX Mechanical Switch”, with seven labelled parts, a Hold to Press button, an explode slider, an X-ray view and an indicator that changed from OPEN//IDLE to ACTUATED//CLOSED.
Honest caveat: I pressed the switch with the Hold to Press button. I did not test clicking the keycap itself, which the prompt asked for.
Test 2: turning a README into an architecture dashboard
I pasted the README of DarshJDB, my self-hosted Rust backend, after this prompt:
Here is the README of DarshJDB, my self-hosted Rust backend on PostgreSQL. Turn it into an interactive architecture dashboard as a single web page: a clickable diagram of the main components and how data flows between them, a side panel that explains whichever component I click, and a short "request lifecycle" walkthrough I can step through. Only use what the README says, and do not invent features. Dark theme.
What happened: after about three and a half minutes it produced a “DarshJDB Architecture Dashboard”: a clickable map of the components (the API on port 7700, the server, the PostgreSQL triple store, the RESP3 cache server on 7701 and more), a detail panel for whichever part I clicked, and a four-step request lifecycle walkthrough. I spot-checked the facts against the README.
Honest caveat: the side panel rendered below the diagram, not beside it as I asked.
Try it with your own project
Swap in your own material. The pattern that worked for me:
- Say exactly what to build and that it must be a single self-contained page.
- List the parts or components you expect to see.
- Name the interactions you want, such as click, slider, step-through.
- For your own docs, paste the source and say “Only use what it says, and do not invent features.”
- Then check the result against your source, as I did. A good-looking page can still be wrong.
What Google says Argon has done
These are Google’s own claims from its launch post, not my tests:
- Argon agents replaced 32K lines of SIMD code in libgav1, Google’s open-source video decoder, with safe Rust. Google says the result runs 2.7x faster than the earlier Rust port, with identical video output.
- A team of Argon agents found and applied memory optimisations across Google’s data centres, freeing over 300 TiB of memory once rolled out, according to Google.
For developers: the API
- Model code:
gemini-4-argon - Context: a 1M-token unified input-output limit via the Gemini API, according to Google. Default output is 65,536 tokens, with long decode up to about 1M.
- Thinking levels: low, medium and high.
- Price per 1M tokens on the standard tier: $2 input and $10 output until 30 November 2026, then $4 and $20 from 1 December 2026. I checked Google’s pricing page on 19 October 2026; check it again before you budget.
Sources & further reading
- Gemini 4 Argon: our next era of frontier intelligence — Google
- Gemini 4 Argon: built to solve your hardest problems — Google
- Gemini 4 Argon is now available in the Gemini app — Google
- Gemini 4 Argon model page — Gemini API docs
- Gemini API pricing — Google AI for Developers
This page is an independent explainer. It is not sponsored by Google and does not imply Google endorsement. Product details are as described by Google at the time of writing. The two tests are my own runs on my own account.
Frequently asked questions
What is Gemini 4 Argon?
Gemini 4 Argon is Google’s frontier Gemini model, announced on 30 September 2026. Google says it is built for complex, multi-step work such as real-world software engineering, enterprise knowledge work and cybersecurity defence.
Who can use Gemini 4 Argon in the Gemini app?
According to Google, Google AI Pro and Ultra subscribers can use it in the Gemini app from 13 October 2026. Developers can use it through the paid tier of the Gemini API; it is not on the free tier.
What is the API model name and price?
The Gemini API model code is gemini-4-argon. Google’s pricing page lists introductory standard-tier prices of $2 per million input tokens and $10 per million output tokens until 30 November 2026, then $4 and $20 from 1 December 2026. Check Google’s pricing page for current rates.
Were the demos in the reel real?
Yes. Both demos were single prompts run in the Gemini app on my own account with 4 Argon selected at the default thinking level, with no follow-ups or edits. The exact prompts are on this page.