How to Get Gemini 4 Argon: Who Can Use It Today, How to Be Next in Line, and What Doesn't Exist Yet

Short answer: as of October 1, there is no way for a US developer or consumer to use Gemini 4 Argon, and no form to join a queue. Google's announcement names three waves — 'a set of trusted cyber defenders through our Fairwind Program' now; then 'paid API customers and Google AI Ultra subscribers'; then 'developers, enterprises, and consumers as soon as possible' — and gives no date for the second or third. We checked every official surface this morning: the Gemini API models and pricing pages, the API changelog, the Gemini app release notes, the Gemini Enterprise Agent Platform model list, Antigravity's docs, and the third-party gateways. No Argon anywhere, no model ID, no waitlist, no 'notify me.' What follows is the honest version of 'what do I do next': one application that exists but almost certainly isn't for you, two things to set up so you're first when the door opens, and a way to tell the real release from the fakes.

Door 1: the Fairwind Program (open, and probably not for you)

The only application connected to Argon is Google DeepMind's Fairwind Program, launched September 2 with the smaller Gemini 3.8 Flash Cyber model. It 'gives high-priority defenders (like governments, healthcare providers, and telecommunications services) early access to advanced models,' and 'a set of Fairwind Program partners get exclusive access to Gemini 4 Argon,' usable inside Google's CodeMender vulnerability tool. Google says it works 'with over 650 partners globally' — note 'a set of,' not all 650 get Argon.

For a typical startup or independent developer, this door is closed. The language is organizations, infrastructure, and background checks, and the use is restricted to security teams doing 'authorized threat simulation, reverse engineering, and malware analysis.' If you run security for a hospital network, a utility, a telecom, or a widely used software platform, apply. If you want Argon to write your app, you're waiting for wave two like everyone else.

Door 2: become a 'paid API customer' now, so you're first in wave two

Google hasn't defined which paid tier gets Argon first, or whether Vertex billing counts the same as AI Studio billing. What it has published is how its API tiers work, on the rate-limits page — and the sensible move is to be a paying customer before the switch flips rather than scrambling after.

• If you've never paid Google for API access: open Google AI Studio, link a billing account, and run something small on Gemini 3.8 Flash — $0.75 per million input tokens through December. That makes you Tier 1, which is the literal definition of 'paid API customer.'
• If you're on Vertex / the Gemini Enterprise Agent Platform: Argon isn't in the model list yet; the latest entry is Claude Sonnet 5.5, added September 28. Google's zero-data-retention policy applies to 'all managed models… including GA and pre-GA models,' so expect Argon to arrive there as a pre-GA managed model.
• Budget for the real price: $2 in / $10 out per million tokens 'during introductory pricing,' per Logan Kilpatrick, rising to $4/$20 'after the introductory period expires' — Google gives no end date. Cached input is 95% off.

Door 3: Google AI Ultra (if you're a Gemini app user)

Consumers get Argon through one plan first: Google AI Ultra, which starts at $99.99 a month for '5x higher usage limits vs. AI Pro' or $199.99 for 20x, and already includes 'first access to advanced features like Deep Think and Gemini Spark.' It's available 'in more than 150 countries' to users over 18, cancel anytime, no partial refunds. What Google hasn't said: whether Argon will appear in Ultra as a model-picker option or inside a Deep Think-style mode, and when. What's clear: the $19.99 AI Pro plan isn't named in the first wave, and the free tier — currently on 3.6 Flash — isn't mentioned at all. If you're on Pro and tempted to upgrade for Argon, wait for a date; $100 a month for a model with no arrival time is a bet, not a purchase.

Why the wait: the government process Google mentioned

Google's post says it is 'actively engaged in the U.S. government's voluntary process for pre-release model access while we gradually expand access.' That's NIST's Center for AI Standards and Innovation, which signed frontier-model testing agreements with Google DeepMind, Microsoft, and xAI in May to run 'pre-deployment evaluations' for cyber risk; it has reportedly completed more than 40 such evaluations. It's voluntary, and no source describes it as able to block a release — Google is choosing to finish it before the broader rollout, which is a reason to expect weeks rather than days, but not a legal gate. (It's separate from the White House accord Sundar Pichai signed the day before the launch, which covers monitoring and auditing, not pre-release access.)

What doesn't exist — so you don't get scammed

• No waitlist or early-access form. Not on blog.google, deepmind.google, or ai.google.dev. The Fairwind form is the only one, and it's for organizations.
• No model ID. 'gemini-4-argon' appears nowhere on Google's pages; the only place we found it is an unmerged pull request in a community routing project, added as metadata with a note that the model is 'not yet generally available.'
• No model card — the URL returns a 404 — and no API documentation.
• No third-party access. Vercel AI Gateway (396 models) tops out at Gemini 3.8; OpenRouter lists 72 Google models, none Argon; nothing on Cursor, Copilot, Poe, or Perplexity. Pre-launch 'Argon' sightings on Arena were anonymous test routes, not access.
• Any site or app claiming to offer Argon today is unofficial at best. If the model isn't on Google's API, nobody can resell it.
• No date. Google's staff said 'as soon as possible' five different ways on launch day; no outlet got a spokesperson to say 'weeks' or 'October.'

While you wait

Argon's introductory price is exactly Claude Sonnet 5.5's and GPT-6.1 Sol's, both available today, and the independent scores so far put Argon level with GPT-6 Astra overall and behind Anthropic's 5.5 models on coding — our Argon breakdown has the three scoreboards. If your work is long-context or agentic office tasks, where Argon's numbers are strongest, Gemini 3.8 Flash at $0.75/$3.75 is the cheapest way to build the integration now and swap the model string later. If it's coding, Opus 5.5 or Sonnet 5.5 are the models to beat regardless of when Argon lands.

How you'll know it's really available

• A row on ai.google.dev/gemini-api/docs/models with a model ID and a price — Google has never shipped a flagship to developers without one.
• An entry in the Gemini API changelog — the last entry is September 22.
• Argon in Google AI Studio's model picker, and in the Gemini Enterprise Agent Platform's model list.
• A line in the Gemini app release notes naming the plan it reaches — the newest entry is September 30's 'Skills,' not Argon.
• We'll update this page and our Gemini review the day any of those happens.

Our read

Google announced a model and opened one door, to the people least likely to be reading this. That's defensible — cyber defenders got a frontier model without guardrails a day after Google's CEO signed a safety accord, which is a coherent story — but it leaves everyone else with a price list for something they can't buy. The practical advice is unglamorous: link a billing account so you're a 'paid API customer' on paper, don't upgrade to Ultra on a promise, build on 3.8 Flash or Sonnet 5.5 at the same price today, and bookmark the four pages above. When Argon shows up on one of them, the switch is a model string.