Gemini 4 Argon: $2 / $10 Introductory Price, Not Yet on the API
Google has announced its next frontier model and its price, but only a small group of security testers can use it so far.
Google announced Gemini 4 Argon on 30 September 2026. It will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input 95% off, which works out at $0.10 per million. For now, it is not available to developers: Google is rolling it out first to "trusted cyber defenders" through its Fairwind Program.
Who can use it, and when
Google says it will gather feedback from early testers and strengthen its safeguards before releasing Argon more widely, "starting with paid API customers and Google AI Ultra subscribers". It has not given a date. Gemini 4 Argon does not yet appear on Google's Gemini API or Vertex AI pricing pages, and has no public API model ID.
Because we only list prices that can be checked on a provider's own pricing page and paid for today, Gemini 4 Argon is not in our price table yet. We will add it when it reaches the paid API.
The announced price in context
| Model | Input / output per 1M | Cached input |
|---|---|---|
| Gemini 4 Argon (announced) | $2.00 / $10.00 | $0.10 |
| Gemini 3.1 Pro | $2.00 / $12.00 | $0.20 |
| GPT-6.1 Sol | $2.00 / $10.00 | $0.10 |
| Claude Sonnet 5.5 | $2.00 / $10.00 | $0.20 |
| Claude Opus 5.5 | $4.00 / $20.00 | $0.20 |
At the introductory price, Argon matches GPT-6.1 Sol exactly and costs less on output than Gemini 3.1 Pro, the current top Gemini model on the API. Google calls the price introductory but has not said how long it will last or what it will cost afterwards, so budget for it to rise.
What Google says it can do
- Very long outputs: Argon can write up to 1M output tokens in a single response, up from 64K on earlier Gemini models. Long reasoning runs are billed as output, so a single hard request could cost several dollars at the introductory rate.
- Coding: 77.9% on DeepSWE v1.1, which Google says is a new state of the art for long-horizon software engineering.
- Knowledge work: top of the Vals Index, which covers finance, coding, legal and tax work, and first on Zapier's AutomationBench at 51.3%.
- Security: finding and patching vulnerabilities, which is why cyber defenders get it first, and Google's strongest resistance yet to prompt injection.
These are Google's own claims. On the independent Artificial Analysis Intelligence Index, which tested it ahead of general release, Argon scored 53 on 2 October, level with GPT-6 Astra, one point above GPT-6.1 Sol and behind Claude Opus 5.5 (58) and Claude Sonnet 5.5 (56).
Should you wait for it?
Not if you need a model now. Gemini 3.1 Pro and Gemini 3.8 Flash remain Google's options on the paid API, and GPT-6.1 Sol and Claude Sonnet 5.5 cost the same per token as Argon's introductory price. If Argon's long-output reasoning suits your work, it is worth testing when it reaches the API. Our cost calculator can estimate a workload in the meantime.
Sources: Google announcement, Gemini API pricing, Artificial Analysis.