Google Announces Gemini 4 Argon: Limited Access First, API Pricing and a 1M Token Output Limit
Google announced Gemini 4 Argon, its new top-tier AI model, on September 30, 2026, but most businesses cannot use it yet. As of October 6, 2026, access is limited to a set of trusted cyber defenders in Google's Fairwind Program, and Google has not given a date for wider release. Google has published API prices and says paid API customers and Google AI Ultra subscribers will be first in line when access opens.
That makes this an announcement to plan around, not a product to switch to this week. If your company already runs Gemini inside a product, or is weighing which model to use in new custom software and AI development, the useful questions are who gets access, what it will cost, and what you can sensibly prepare before it arrives.
What did Google announce on September 30, 2026?
Google announced Gemini 4 Argon, which it describes as a frontier model built for complex, long-running work. The announcement on Google's blog names three target areas: real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense.
Two details stand out for people who buy software and not just people who write it. First, Google is raising the model's output limit to 1 million tokens, up from a previous limit of 64K tokens. Second, Google is releasing the model in phases, starting with a restricted security group, and says that "safely releasing frontier capabilities at this level requires a phased approach".
Who can use Gemini 4 Argon right now?
Only a limited group can use it today. Google says Argon is "rolling out to a set of trusted cyber defenders through our Fairwind Program", which is its route for giving security teams early access before a general release.
Google also says it is "actively engaged in the U.S. government's voluntary process for pre-release model access". Businesses, developers and consumers outside those groups do not have access yet. When checked on October 6, 2026, the Gemini API changelog had no entry for Argon and the Gemini API pricing page did not list it, which is consistent with a model that has been announced but not opened to API customers.
When will businesses get access to Gemini 4 Argon?
Google has not announced a date. Its stated plan is to gather feedback from early testers while it adjusts guardrails, then make Argon available to developers, enterprises and consumers "as soon as possible".
The order of release is clearer than the timing. Google says the wider rollout will start with paid API customers and Google AI Ultra subscribers. If your company uses the Gemini API on a paid plan, you are in the first group. If you use Gemini on a free tier or through another subscription level, expect to wait longer, and treat any specific date you read elsewhere as unconfirmed until Google publishes it.
How much will Gemini 4 Argon cost?
Google has stated two sets of API prices: an introductory rate and a higher standard rate. The introductory rate is USD 2 per million input tokens and USD 10 per million output tokens, with cached input tokens priced at 95 percent off. After the introductory period, the price becomes USD 4 per million input tokens and USD 20 per million output tokens.
The announcement does not say how long the introductory period lasts. That gap matters for budgeting, because the standard rate is exactly double.
An illustrative calculation
Suppose one task sends 200,000 input tokens (a large set of documents) and receives 50,000 output tokens (a long report). At the introductory rate that is USD 0.40 for input plus USD 0.50 for output, or USD 0.90. At the standard rate the same task costs USD 0.80 plus USD 1.00, or USD 1.80. This example is illustrative and ignores caching discounts. The safe approach is to build your business case on the standard rate and treat the introductory rate as a temporary saving.
What does a 1 million token output limit mean in practice?
It means the model can return a far longer response in one request than before. A token is a piece of a word, and the output limit caps how much a model can write back in a single reply. Moving from 64K to 1 million tokens is roughly a sixteen-fold increase in that cap.
In practical terms, long outputs that previously had to be split into many requests and stitched together, such as a full technical document, a large code change across many files, or a converted data set, could be produced in one pass. That removes some engineering work.
It also creates a cost risk. Output tokens are the expensive side of the price list, and a reply that used the whole limit would cost USD 10 at the introductory rate and USD 20 at the standard rate for the output alone. Any application that uses Argon should set its own maximum output size per request instead of leaving the full limit open.
How should a business read Google's benchmark claims?
Treat them as the vendor's own figures until your team has tested the model on your work. Google reports, among other results, a score of 77.9 percent on the DeepSWE v1.1 software engineering benchmark and a first-place 51.3 percent on AutomationBench.
Benchmarks measure a fixed set of tasks under the vendor's test conditions. They do not tell you how a model handles your documents, your codebase or your customers' questions, and a score of 51.3 percent on an automation benchmark is also a reminder that even a leading model fails a large share of hard multi-step tasks. For any process where mistakes are costly, human review and clear permissions still matter more than which model ranks first. Our guide to what an AI agent is explains those controls.
What should a business do now?
A business should prepare to evaluate Argon, not commit to it. Nothing in the announcement requires action, and no existing Gemini model is being withdrawn as part of it. These steps are worth doing while access is closed:
- Confirm whether your company is a paid Gemini API customer or has Google AI Ultra, since those groups get access first.
- Write down the three to five tasks where your current model falls short, with real examples, so you have a test set ready on the first day.
- Estimate costs using the standard rate of USD 4 and USD 20 per million tokens, not the introductory rate.
- Ask your developers whether the model name in your application is a setting that can be changed without rewriting code.
- Decide in advance what result would justify switching, such as fewer errors on a specific task or fewer requests per job.
Do not delay a project that works with a model available today on the expectation that Argon arrives soon. Google has given no date, and a well-built application can change models later.
Quick answers
When was Gemini 4 Argon announced?
Google announced Gemini 4 Argon on September 30, 2026. It is described as a frontier model for software engineering, enterprise knowledge work and cybersecurity defense.
Can my business use Gemini 4 Argon today?
Probably not. As of October 6, 2026, Google says Gemini 4 Argon is rolling out only to a set of trusted cyber defenders through its Fairwind Program, and it has not given a date for wider access.
Who will get Gemini 4 Argon first when access widens?
Google says the wider rollout of Gemini 4 Argon will start with paid API customers and Google AI Ultra subscribers. No date had been announced as of October 6, 2026.
How much does Gemini 4 Argon cost in the API?
Google states an introductory price of USD 2 per million input tokens and USD 10 per million output tokens for Gemini 4 Argon. After the introductory period the price is USD 4 per million input tokens and USD 20 per million output tokens; the length of that period has not been stated.
What is the output limit of Gemini 4 Argon?
Google says Gemini 4 Argon has an output limit of 1 million tokens, up from a previous limit of 64K tokens. This lets the model return much longer responses in a single request.
Should I wait for Gemini 4 Argon before starting an AI project?
No. Google has not announced a general release date for Gemini 4 Argon, and an application built so that the model is a changeable setting can switch later. Start with a model that is available now and test Argon when you have access.
Conclusion
Gemini 4 Argon was announced on September 30, 2026 with a 1 million token output limit and published API prices, but as of October 6, 2026 it is available only to a restricted group of cyber defenders. Paid API customers and Google AI Ultra subscribers are next, with no date given. The sensible response is to prepare a test set, budget on the standard price, and make sure your software can change models without a rebuild.
Entrant Technologies builds websites, web applications, mobile apps and custom software. If you want help deciding how a new model fits into software you are planning or already run, you can contact us and we will take a look.