Gemini 4 Argon Is Out of Testing: Where to Use It and What It Costs

Executive Summary

• What happened: On October 13, Google made Gemini 4 Argon available beyond its first testers. It was announced on September 30 for trusted cyber defenders only.
• Who gets it: developers on the paid Gemini API tier (AI Studio, Android Studio), Google AI Ultra subscribers in Antigravity, Pro and Ultra subscribers in the Gemini app and AI Mode in Search, and businesses through Gemini for business. It is not on the free tier.
• Price: $2 per 1M input tokens and $10 per 1M output tokens until November 30, then $4 and $20. Cached input is 95% off.
• How good: #1 on the independent Vals Index (68.9%), but 53 on the Artificial Analysis index, five points behind Claude Opus 5.5.
• For developers: the model ID is gemini-4-argon, with a 1M-token limit shared by input and output.

I made a short video of this on X; this post has the details and a first API call.

Gemini 4 Argon is open

From testers to a global rollout

When Google announced Argon on September 30, only members of its Fairwind Program, trusted cyber defenders, could use it. On October 13 the rollout opened up. Here is who gets it where, from Google's announcement and the developer post:

• Gemini API: the paid tier, through Google AI Studio and Android Studio.
• Google AI Studio and Antigravity: Google AI Ultra subscribers can try it there too.
• Gemini app and AI Mode in Search: Google AI Pro and Ultra subscribers (Gemini app post).
• Businesses: Gemini for business, the API and Antigravity.
• Cyber defenders: a version without cyber guardrails through the Fairwind Program, on a rolling basis.
• Scientists: Google is offering 10,000 researchers six months of Antigravity's highest usage tier for free, subject to selection.

So "available for everyone" means everyone who pays. If you are on the free Gemini API tier or a free Gemini account, you don't get Argon yet.

Illustration: Argon's products unlock on October 13, each with who can use it

The price, and the date to remember

Google launched Argon at an introductory price and wrote the end date into the announcement's footnote:

• Until November 30, 2026: $2 per 1M input tokens, $10 per 1M output tokens.
• From December 1, 2026: $4 per 1M input tokens, $20 per 1M output tokens.
• Cached input: 95% off the input price, both before and after.

If you build a cost estimate this month, use the December numbers. A workload that costs $100 a day in November costs about $200 a day in December. Artificial Analysis also notes Argon is a verbose model (110M output tokens on its index runs against a median of 85M), and output tokens are the expensive side.

Illustration: the intro price rolls to the standard price on December 1

How it scores on independent tests

Google's own charts show Argon leading on several tests. Two independent leaderboards give a more mixed picture:

• Vals Index: 68.9% on Vals AI, first of 41 models, ahead of Claude Sonnet 5.5 (67.0%) and Claude Opus 5.5 (67.0%). It covers finance, coding, legal and tax tasks.
• Intelligence Index: 53 on Artificial Analysis, #8 of 216, tied with GPT-6 Astra and Claude Fable 5.1 and behind Claude Opus 5.5 at 58 (VentureBeat).

Google also reports 87% on the ERQA embodied reasoning test and a 0.7% attack success rate on the Gray Swan indirect prompt injection benchmark. Those are Google's numbers. My read: Argon is a top-tier model for long knowledge-work tasks, not a clear #1 everywhere, so test it on your own workload before you switch.

Illustration: #1 on the Vals Index, behind Claude Opus 5.5 on Artificial Analysis

What Google says it already did with it

Google has been using Argon internally, and the announcement lists results from its own teams:

• Antigravity at Google: agentic code submissions up more than 35%, with code rollbacks down by nearly a third.
• Chrome: an architectural refactor across hundreds of files, separating core webpage handling from a ~6 million line legacy folder, done in weeks instead of months.
• Gemma inference: a 2.85x end-to-end speedup on an A100 with vLLM, from about 37,900 to over 108,000 tokens per second on Gemma 4 E2B, with 100% token parity.
• Agent Development Kit: working overnight on its own, it found changes that cut in-memory session prep time by 40% in long sessions and made cold imports 21% faster.

These are Google's own projects and measurements, so treat them as examples of what the model can do rather than a guarantee.

Google's numbers: 35%+ more agent code submissions, 2.85x faster Gemma inference, 40% less ADK session prep

Your first API call

The model page lists the ID gemini-4-argon as stable, with a 1,048,576-token context window shared by input and output, 65,536 output tokens by default for a normal request and up to the full 1M with long decode. It takes text, images, video, audio and PDFs and returns text. Function calling, structured outputs, code execution, search grounding and URL context are supported.

Google's current SDK samples use the Interactions API. Install the SDK and set your key:

pip install -U google-genai
export GEMINI_API_KEY="your-paid-tier-key"

Then swap Argon into the quickstart call:

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Refactor this module. Keep tests green.",
)
print(interaction.output_text)

Argon thinks at a medium level by default. For hard multi-step work you can raise it, as the thinking docs show (low, medium or high):

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Plan the migration of this service to the new queue.",
    generation_config={"thinking_level": "high"},
)

Higher thinking means more output tokens, so check the cost against the December price. For very long outputs, read the docs on the new continuation protocol before you rely on the 1M limit.

Illustration: the first call with the gemini-4-argon model ID

Sources

• Gemini 4 Argon is now available for everyone (Google, October 13, 2026)
• Gemini 4 Argon: Built to solve your hardest problems (Google DeepMind)
• Google DeepMind on X (launch post)
• Gemini 4 Argon: our next era of frontier intelligence (Google, September 30, 2026)
• Gemini 4 Argon is now available for developers and in the Gemini app (Google)
• Gemini API docs: model page, quickstart, thinking
• Vals AI: Gemini 4 Argon and Artificial Analysis: Gemini 4 Argon
• VentureBeat and Reuters via US News