Gemini 4 Argon: what Google's new model is, what it costs, how to use it

The short answer: Gemini 4 Argon is Google's new flagship AI model, built for long, careful reasoning on hard problems such as coding, maths, science, finance and multi-step agents. It reads up to about 1 million tokens, can write up to 1 million tokens in a single answer, and costs $2 per million input tokens and $10 per million output tokens until 30 November 2026 (then $4 and $20). Developers can use it now in the Gemini API and AI Studio, and Google AI Pro and Ultra subscribers can use it in the Gemini app.

Google announced Argon on 30 September and made it widely available on 13 October 2026. Below are the facts that matter, where you can use it, what it costs, and how to call it, with the sources linked.

Gemini 4 Argon at a glance

Sources: Google's announcement and the Gemini API model page.

What is Gemini 4 Argon?

Gemini 4 Argon is the first model in Google's Gemini 4 family and its new top reasoning model. Google describes it as built to keep reasoning steadily through complex questions, and aims it at software engineering, maths and science, knowledge work, cybersecurity defence, creative writing, and autonomous agents that run many steps on their own. Under the hood it is a large sparse mixture-of-experts transformer, according to Google DeepMind's model card.

What's new compared with earlier Gemini models?

• Very long answers. The output limit can now go up to about 1 million tokens with "long decode", up from the previous 64K. That matters for agents writing whole codebases, long reports, or big data transformations in one go.
• Adjustable thinking. You choose low, medium or high thinking, trading speed and cost for depth.
• Agent tools built in. Function calling, code execution, Google Search grounding, file search, structured outputs, and computer use (in preview).
• Safety work aimed at agents. Google reports a 0.7% success rate for indirect prompt-injection attacks in Gray Swan's testing, and about half as many misaligned-behaviour violations as its earlier models.

How good is it?

Two independent results stand out:

• Vals Index: Argon ranks first on the Vals AI index, which tests real work in finance, coding, law and tax.
• Hallucinations and cost: Engadget reports that Argon matches OpenAI's GPT-6 Astra on an overall intelligence index at about 60% of the cost, with a 15% hallucination rate against 54% for GPT-6 Astra and GPT-6.1 Sol.

Google also reports 87% on ERQA (spatial and physical reasoning) and 52.2% on LabBench (wet-lab reasoning). Benchmarks are a starting point, not a verdict. The real test is your own tasks.

How much does Gemini 4 Argon cost?

Cached input tokens are 95% cheaper. For comparison, Engadget lists GPT-6 Astra at $10 input and $50 output per million tokens. If you're going to try Argon on a big job, the introductory window is the cheap time to do it.

Where can you use it?

• Developers: the Gemini API and Google AI Studio (paid tier).
• Companies: Google Cloud's Gemini platform and Gemini for business.
• Everyone else: the Gemini app and AI Mode in Google Search for Google AI Pro and Ultra subscribers.
• Coding: Google Antigravity, for Google AI Ultra subscribers.
• Scientists and security teams: Google is giving scientists free top-tier access, and cybersecurity defenders get it through its Fairwind Program.

How do I use Gemini 4 Argon in code?

Use the model ID gemini-4-argon with Google's Gen AI SDK and an API key from AI Studio:

from google import genai

client = genai.Client()  # reads GEMINI_API_KEY from your environment

response = client.models.generate_content(
    model="gemini-4-argon",
    contents="Review this function for bugs and explain each fix: ...",
)
print(response.text)

pip install google-genai

Start with medium thinking, and move to high only for the tasks that need it. High thinking costs more output tokens.

Should you switch to Gemini 4 Argon?

My take, as someone who builds AI agents and automations for businesses:

• Try it if your work is long and reasoning-heavy: big code reviews, research reports, agents that run for many steps, or anything that needs a very long answer in one go. The 1M-token output is unusual.
• Run it on your real tasks first. I pick models per job, not per leaderboard. Run five of your own tasks through Argon and the model you use today, and compare the quality, the cost and the number of times you had to fix something.
• Use the cheap window. At $2/$10 until 30 November, testing it now costs half of what it will later.

I haven't run Argon on my own client builds yet. When I do, I'll write up the results here.

FAQ

Is Gemini 4 Argon free?

Not for the API: it's on the paid tier at $2/$10 per million tokens until 30 November 2026. Regular users get it through Google AI Pro and Ultra subscriptions in the Gemini app and AI Mode in Search. Google gives scientists free top-tier access.

What is the context window of Gemini 4 Argon?

1,048,576 tokens of input. Output is 65,536 tokens by default and up to 1,048,576 tokens with long decode.

When was Gemini 4 Argon released?

Google announced it on 30 September 2026 and made it generally available to developers and businesses on 13 October 2026.

Is Gemini 4 Argon better than GPT-6 Astra?

On the numbers published so far, it matches GPT-6 Astra on an overall intelligence index at about 60% of the cost, with far fewer hallucinations (15% vs 54%). It ranks first on the Vals index. Which one is better for you depends on your tasks, so test both.

What's the Gemini 4 Argon model ID?

gemini-4-argon.

Want an AI agent or automation built on the right model for your business? Open a case file. I reply within 24 hours.