Gemini 4 Argon: 1M Output Tokens, $2 Input — and You Can't Use It Yet
Google's new frontier model writes up to 1 million tokens in one answer and launches at $2 per million input tokens. It is also, for now, only available to a hand-picked group of cyber defenders.
यह लेख हिंदी में पढ़ें
Google announced Gemini 4 Argon on 30 September as its next frontier model, aimed at long software-engineering tasks, legal and financial knowledge work, and cyber defence. The headline numbers are striking. The release is not: for now, Argon is rolling out only to "a set of trusted cyber defenders" through Google's Fairwind Program.
The 1 million figure is output, not context
Several write-ups describe Argon as having a million-token context window. That is not what Google says. Its announcement describes an "industry-leading 1M tokens" of output, "up from the previous 64K". That is a much more unusual claim: it means Argon can produce an extremely long piece of work — a large code migration, a long report — in one response, rather than in many stitched-together calls.
Google's post does not state Argon's input context window. Until it does, treat any number you see for it as unconfirmed.
What it costs — when you can get it
| Input (per 1M tokens) | Output (per 1M tokens) | |
|---|---|---|
| Introductory price | $2 | $10 |
| Standard price (after the introductory period) | $4 | $20 |
| Cached input | 95% off the input price | — |
At 95% off, cached input works out to $0.10 per million at the introductory rate and $0.20 at the standard one. Google has not said how long the introductory period lasts. For comparison, the two flagships released in September — GPT-6 Astra and Claude Fable 5.1 — both list at $10 input and $50 output per million. On paper Argon is a fraction of that. In practice you cannot buy it yet.
The benchmarks Google published
These are Google's own figures from the announcement. Google ran them; no independent lab has replicated them yet.
| Benchmark | What it measures | Argon's score |
|---|---|---|
| DeepSWE v1.1 | Real-world software engineering | 77.9% |
| AutomationBench | Automating multi-step tasks | 51.3% (ranked #1) |
| LVBench | Long-video understanding | 91.7% |
| CWE-bench v1 | Finding and fixing security weaknesses | 68% (tied for first) |
Google also says Argon leads on the Vals Index, the Gray Swan prompt-injection benchmark, Vals Finance Agent v2 and Harvey's Legal Agent Benchmark, but does not give scores for those. A claim of leading without a number is worth less than one with a number, so we have left those out of the table.
Why the release is so narrow
Google frames it as safety. Its announcement says releasing frontier capability "at this level requires a phased approach", citing defence against misuse, prompt-injection attacks, model misalignment and system vulnerabilities. It is also taking part in the US government's voluntary process for pre-release model access.
The choice of first users is telling. A model this good at finding and patching security flaws is, by the same skill, good at finding them to exploit. Giving it to defenders first is a way of letting them patch before the capability spreads.
Who gets it next, and what that means for India
- Now: selected cyber-security teams in Google's Fairwind Program.
- Next, per Google: paid Gemini API customers and Google AI Ultra subscribers, "as soon as possible".
- Timeline: not given. Google's post does not mention India or any other country.
For an Indian developer the practical answer today is simple: you cannot call Argon, so there is nothing to migrate to. If your workload is long code migrations or long-form generation, the 1M output limit and the introductory price are worth planning around once access opens — and worth checking against the AI cost calculator with real token counts rather than the headline rate.
Sources
- Google: Gemini 4 Argon — our next era of frontier intelligence (announcement, 30 September 2026)
Frequently asked questions
Can I use Gemini 4 Argon in India right now?
No. As of its 30 September 2026 announcement, Argon is rolling out only to selected cyber-security teams in Google's Fairwind Program. Google says paid API customers and Google AI Ultra subscribers come next, but has not given a date or said which countries.
Does Gemini 4 Argon have a 1 million token context window?
Google's announcement says 1 million output tokens, up from 64K — the length of answer it can produce. It does not state the input context window, so a 1M context figure is not confirmed by Google.
How much does Gemini 4 Argon cost?
Introductory pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Cached input is 95% off the input price. Google has not said how long the introductory period lasts.
Save this summary as an image or share it.
AICreatorHub Team
The AICreatorHub editorial team is a group of hands-on AI practitioners, writers and developers based in India. We test AI tools and models ourselves, track official releases from OpenAI, Anthropic, Google, Meta and xAI, and translate them into simple, India-first guides in English and Hindi. Every article is written for real Indian use cases — pricing in rupees, free-tier tips and practical, tested steps — so you get accurate, up-to-date and genuinely useful AI information.