Gemini 4 Argon is Google’s new top model, announced on September 30, 2026, and almost nobody outside a cybersecurity program can use it yet. Google is releasing it first to "a set of trusted cyber defenders" through its Fairwind Program. Developers, businesses and Gemini app users come later, with no date given.
This piece covers what Google announced, who has access as of October 7, what Fairwind is, the published prices, and where Argon fits in the Gemini 3.x lineup. It also compares Argon with the current top models from OpenAI and Anthropic, and ends with what agencies and small organizations should do this week.
The short version: Argon is real, it is priced, and it is not in the Gemini API, the Gemini app or Google Cloud yet. Google says the first wider release will go to paid API customers and Google AI Ultra subscribers. When it arrives, the introductory price matches Claude Sonnet 5.5 and GPT-6.1 Sol per token, and its 1M-token output limit is far above anyone else’s. For now, keep shipping client work on the models you have.
What Google announced with Gemini 4 Argon
Google’s Gemini 4 Argon announcement is written by Koray Kavukcuoglu of Google DeepMind. It says Argon is "built to sustain deep reasoning across complex, long-horizon workflows." Google names three strengths: software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense.
| Item | What Google says | Status |
|---|---|---|
| Announced | September 30, 2026 | Confirmed |
| Output limit | 1M tokens, up from 64K | Confirmed |
| Context window | Not stated | Unknown |
| Price | $2 in, $10 out per million tokens (introductory); $4 and $20 after | Published, not yet billable |
| Model ID | None published | Not in the API docs |
| Who has it | Fairwind Program cyber defenders | Phased rollout |
Google’s own benchmark figures include 77.9% on DeepSWE v1.1, a long-horizon software engineering test, and 51.3% on Zapier’s AutomationBench. On CWE-bench v1, which tests fixing security flaws, Google says Argon ties for first at 68%. These are Google’s numbers, so treat them as a starting point.
Who can use Gemini 4 Argon today
As of October 7, the answer is Fairwind members only. We checked every Google page where a release would normally show up.
| Where | Argon listed? | What we found |
|---|---|---|
| Gemini API changelog | No | Newest entry October 6 (Nano Banana 2.1); no Argon |
| Gemini API models and pricing pages | No | Newest Pro is still the 3.1 Pro preview |
| Gemini app release notes | No | September 30 entry covers Skills, not Argon |
| Google Cloud release notes | No | No Argon entry through October 5 |
| Fairwind Program | Yes | “Exclusive access to Gemini 4 Argon” |
Google’s September AI recap, posted October 2, says the same thing. Google is "actively gathering feedback from early testers to iterate on guardrails" before it expands access to developers, enterprises and consumers.
The announcement names who comes next: "paid API customers and Google AI Ultra subscribers." It does not mention Workspace, Gemini Enterprise or Antigravity. That points to API access before Gmail or Docs, but Google has not confirmed any order beyond that sentence.
What the Fairwind Program is
Fairwind is not new with Argon. Google introduced it on September 2, with the Gemini 3.8 Flash and 3.8 Flash Cyber launch. Its first job was giving trusted defenders access to 3.8 Flash Cyber, a model built for finding and patching vulnerabilities.
The Fairwind Program page says it gives "trusted partners a critical head start against cyber threats." It lists three priority groups:
- Governments and national cyber authorities
- Critical infrastructure operators, such as healthcare, telecommunications, energy and financial networks
- Core technology platforms
Applicants need "a proven track record of ethical operations and research," and Google runs background checks. The page says the program has over 650 partners. There is an interest form, and Google says it will respond to eligible partners "as soon as we can." Members can use Argon on its own or with CodeMender, Google’s code security agent. Google also says Wiz is already using Argon in its Scan for Good initiative.
For most agencies and small nonprofits, Fairwind is not a route to Argon. It is a security program, and it screens for that. A hospital system or utility client with a security team might qualify. A marketing site or member portal team will not.
What Gemini 4 Argon will cost
Google published prices before anyone can be billed. The introductory rate is $2 per million input tokens and $10 per million output tokens. Cached input is 95% off, which works out to $0.10 per million. After the introductory period ends, the price becomes $4 input and $20 output.
Google has not said how long the introductory period lasts. It has not said whether long prompts cost more, as they do on the 3.1 Pro preview. It has also not said whether there will be a free tier.
The intro price is lower than the current Pro model on output. The Gemini API pricing page lists the 3.1 Pro preview at $2 input and $12 output for prompts up to 200,000 tokens. Google ran a similar introductory deal on Flash. We covered how the Gemini 3.7 Flash headline price expires on December 31, and any quote built on Argon’s intro rate needs the $4 and $20 figures in it too.
Where Gemini 4 Argon sits in the Gemini 3.x lineup
Google has not said what Argon replaces. Nothing in the Gemini 3.x lineup has a new shutdown date because of it.
As of October 1, the Gemini API models page lists gemini-3.8-flash as stable and gemini-3.1-pro-preview as the newest Pro model, still in preview. The deprecations page shows no shutdown date for the 3.1 Pro preview. Gemini 3.5 Pro, which press reports had expected in June, still does not appear.
Argon also breaks the naming pattern. It is not called Gemini 4 Pro, and Google has not said whether Flash or Flash-Lite versions of Gemini 4 are coming. Until Google says more, think of it as a new top tier above the 3.x models.
For day-to-day work, 3.8 Flash stays the default text model. It has its own costs to watch: our test of what Gemini 3.8 Flash costs per task found it writes far more output than 3.7 Flash. Teams still clearing older IDs should also check our list of Gemini shutdown dates this fall.
How Gemini 4 Argon compares with GPT-6 and Claude 5.5
The comparison below uses list prices from each vendor’s model docs, read October 6. The index column is the Vals Index, version 2.1, dated October 2. Vals is a third-party benchmark that scores agentic work in finance, coding, legal and tax.
| Model | Price per million tokens (in / out) | Max output | Vals Index |
|---|---|---|---|
| Gemini 4 Argon | $2 / $10 intro; $4 / $20 after | 1M | 68.90% |
| Claude Sonnet 5.5 | $2 / $10 | 128K | 67.04% |
| Claude Opus 5.5 | $4 / $20 | 128K | 66.97% |
| GPT-6 Astra | $10 / $50 | 128K | 63.13% |
| GPT-6.1 Sol | $2 / $10 | 128K | 61.15% |
Argon leads the Vals Index, but by less than two points over Claude Sonnet 5.5. Artificial Analysis, whose index we have cited before, did not list Argon when we checked. Argon’s intro price matches Sonnet 5.5 and GPT-6.1 Sol. Its later price matches Claude Opus 5.5.
The other two vendors had a busy week too. On September 28, OpenAI pulled the planned October release of GPT-6.1 Astra, which we covered in OpenAI pauses GPT-6.1 Astra. GPT-6 Astra stays available. OpenAI’s API changelog shows gpt-6.1-sol released on September 29. Anthropic shipped Claude Sonnet 5.5 on September 28, with breaking API changes.
The practical gap is access. GPT-6.1 Sol, Claude Sonnet 5.5 and Claude Opus 5.5 can be called today by anyone with an API key. Gemini 4 Argon cannot.
What a 1M-token output limit is good for
The output limit is the one spec where Argon stands apart. OpenAI and Anthropic cap their current top models at 128K output tokens. Gemini 3.8 Flash and the 3.1 Pro preview stop at 65,536.
A larger output limit matters when one request has to produce a lot: a full set of migration scripts, a long report, or a large batch of rewritten pages. Today those jobs get split into many calls, and your code stitches the parts together.
It is also a cost ceiling to think about. At the intro rate, a single response that used the full 1M output tokens would cost $10. At the later rate it would cost $20. Set your own output limits per request before you let an agent run long.
Should you do anything about Gemini 4 Argon this week?
For most agencies and small organizations, no. There is nothing to switch to. These steps are cheap, though:
- Leave client production work where it is. Gemini 3.8 Flash, Claude Sonnet 5.5 and GPT-6.1 Sol are all available now.
- Keep the model ID in one setting in each site, plugin or script. When Argon reaches the API, testing it becomes a config change.
- Hold off on promises. Do not tell a client their Workspace or Gemini app will get Argon on a set date. Google has not given one.
- Mark the price change in any estimate. If you quote a future project on Argon, show both the intro and the later rate.
- Pass Fairwind to the right clients. A healthcare, energy or financial client with a security team can look at the Fairwind interest form.
Watch the Gemini API changelog for a model ID. That is where an API release would show up first. Our look at security-specialized AI models covers why vendors keep giving defenders early access to models like this one.
Frequently Asked Questions
What is Gemini 4 Argon?
Gemini 4 Argon is Google’s new frontier model, announced September 30, 2026. Google says it is built for long, complex work in software engineering, legal and finance knowledge work, and cybersecurity defense, with a 1M-token output limit.
Can I use Gemini 4 Argon now?
Not unless your organization is in Google’s Fairwind Program. As of October 7, Argon is not listed in the Gemini API, the Gemini app release notes or Google Cloud’s release notes.
Who gets Gemini 4 Argon next?
Google says the wider release starts with paid API customers and Google AI Ultra subscribers. It has not given a date or said when Workspace or other plans will get it.
What is the Fairwind Program?
A Google security program that gives trusted defenders early access to cyber-capable models, first Gemini 3.8 Flash Cyber and now Argon. It prioritizes governments, critical infrastructure operators and core technology platforms, and it screens applicants.
How much will Gemini 4 Argon cost?
Google lists an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After the introductory period, it rises to $4 and $20. Google has not said when that change happens.
What is the Gemini 4 Argon model ID?
Google has not published one. No Argon model appears on the Gemini API models page, pricing page or changelog as of October 7, 2026.
Does Gemini 4 Argon replace Gemini 3.1 Pro or 3.8 Flash?
Google has not said. The 3.1 Pro preview has no shutdown date, and 3.8 Flash is still the stable default for text work in the Gemini API.
How does Gemini 4 Argon compare with Claude and GPT-6?
On the Vals Index dated October 2, Argon scores 68.90%, against 67.04% for Claude Sonnet 5.5, 66.97% for Claude Opus 5.5, 63.13% for GPT-6 Astra and 61.15% for GPT-6.1 Sol. Its 1M-token output limit is about eight times the 128K cap on those models.
Should my agency plan a client project around Gemini 4 Argon?
Not yet. There is no public access, no model ID and no date. Build on models you can call today, keep the model ID in one setting, and test Argon when it reaches the API.
