Skip to content

AI·News & analysis

Google's Gemini 4 Argon reaches the Gemini app, Search and developers

Two weeks after a defenders-only debut, Google is rolling out Gemini 4 Argon to AI Pro and Ultra subscribers and to developers, with a robotics showcase on the side.

By Dan Kost aka Poseidan8 min readX
A white Google sign with the company's colorful logo and the address 1565 and 1585 Charleston Road, standing on wood chips in front of tall redwood trees in Mountain View, California
Photo: Dietmar Rabich / Wikimedia Commons, CC BY-SA 4.0

Tide

Wave

Sci-fi

6/10

Reality

Shipping

A chatbot that rehearses robot chores in a simulation first.How we rate

The Squeeze

Google is rolling out Gemini 4 Argon, its newest frontier AI model, to Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search, and to developers through the paid Gemini API.

Google says Argon answers more directly, with less hedging and fewer cliches. Until November 30 it costs developers $2 per million input tokens and $10 per million output tokens. After that, the price doubles, and free and AI Plus users are still waiting.

What to know

  1. Google is rolling out Gemini 4 Argon to Google AI Pro and AI Ultra subscribers in the Gemini app and in AI Mode in Search.
  2. Google says Argon answers more directly, with less hedging, less repetition and fewer cliches, and reports its progress more often on long tasks.
  3. Developers get Argon in the paid Gemini API at an introductory $2 per million input tokens and $10 per million output tokens until November 30, then double.
  4. Google DeepMind also showed Argon controlling robots, scoring 87% on the ERQA physical reasoning benchmark, which it says beats every other frontier model.

Google announced on Tuesday that Gemini 4 Argon, its newest AI model, is rolling out in the Gemini app. Two weeks ago, the same model was available only to a small group of vetted security experts.

So what changed, who gets it, and what does a chatbot model have to do with a robot dog taking out the trash?

Wait, didn't Argon already launch?

Sort of. Google first announced Gemini 4 Argon on September 30. Back then, it went only to trusted cyber defenders through a vetting program called Fairwind, because the model is very good at finding security holes. We covered that defenders-only debut at the time.

At that point, Google said the model would reach developers, businesses and consumers "as soon as possible," starting with paying customers. Today is that step.

Think of it like a new restaurant that ran a private tasting for critics first. Now the doors are open, as long as you have a reservation, which in this case is a paid plan.

Who can use it now?

Here's where Argon is showing up, according to Google and to reporting by 9to5Google:

  • Gemini app: Google AI Pro and AI Ultra subscribers.
  • AI Mode in Search: also AI Pro and AI Ultra subscribers.
  • Gemini API: developers on the paid tier, through Google AI Studio and Android Studio.
  • Google Antigravity: the desktop app and command line tool require AI Ultra.

Engadget reports that Google AI Pro costs $20 a month in the US, and Ultra costs at least $100.

The catch: free users and Google AI Plus subscribers don't get Argon yet, Android Police notes. The rollout also isn't instant. The Android Police reviewer, a Pro subscriber, still didn't see Argon in the app when the story ran.

In real life If you pay for Google AI Pro, update the Gemini app and tap the model menu next to the text box. When Argon appears there, pick it, then ask the same question you asked yesterday and compare the answers.

9to5Google reports the model picker now lists four options for paying users: 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro and 4 Argon.

What will feel different in the app?

Google's pitch is mostly about style. The company says Argon "answers questions more directly with less hedging, doesn't repeat itself more than necessary, and skips clichés throughout the responses."

In plain terms: fewer answers that open with a compliment, explain three caveats, and then repeat the question back to you. 9to5Google also quotes Google saying Argon "follows style instructions more reliably than any of our previous Gemini models." So if you ask for three bullet points, you should get three bullet points.

For longer jobs, Google says Argon "keeps you in the loop by sharing its progress more often." Engadget adds that it can show responses while it's still thinking, so you aren't staring at a spinner.

Why it matters: most people never read benchmark charts. They notice when an answer is shorter, clearer and closer to what they asked for.

What else is new in Gemini?

Google bundled a few app changes with the model:

  • A tasks toggle: a switch in the main chat window moves you from chat to tasks. Google says Gemini can then work for you "even when you're not at your computer." 9to5Google reports this feature was previously called Spark.
  • More connected apps: Gemini now connects to more than 40 apps, including Instacart, OpenTable, Adobe and Shopify, for things like ordering groceries and booking tables.
  • Coding on the Mac: the agentic coding interface of Google Antigravity, Google's AI development platform, is now built into the Gemini macOS app. Agentic means the AI writes and runs code in steps instead of just suggesting it.

What does it cost developers?

Google's developer announcement calls Argon a model for long, multi-step work like code migrations and research. Prices are per million tokens, the small chunks of text a model reads and writes:

Until November 30, 2026 After November 30, 2026
Input, per million tokens $2 $4
Output, per million tokens $10 $20

Argon's introductory price is half of what it will cost from December, so developers have about seven weeks to test it cheaply.

By the numbers: Engadget cites Artificial Analysis, a company that benchmarks AI models. It says Argon matches OpenAI's GPT-6 Astra on its Intelligence Index, a combined score across many tests, at 60% of the cost per task at the introductory prices.

Google also says it will give 10,000 scientists the highest usage tier of Google Antigravity free for 6 months, 9to5Google reports. Applications are open.

What has Google built with it?

Google shared examples from its own teams using Argon inside an internal version of Antigravity, as quoted by 9to5Google:

  • Chrome: Argon helped separate core webpage handling from a roughly 6 million line legacy folder inside Chromium's 40 million line codebase. Google says that finished "months of work in weeks."
  • Gemma: Argon sped up inference for the small Gemma 4 E2B model by 2.85 times on an Nvidia A100 chip, from about 37,900 to over 108,000 tokens per second.
  • YouTube: Argon cut the CPU use of the comments frontend in half. It found that redundant emoji rendering caused nearly 45% of the server overhead.

That last one is a fun detail. Somewhere, a tiny thumbs-up emoji was quietly working overtime.

Wait, what about robots?

Google DeepMind published a separate post on Argon for robotics. It calls the model "a step change in physical understanding" and says it scores 87% on the ERQA benchmark, a test of solving physical and spatial problems in 3D. DeepMind says that beats every other frontier model.

The demos read like a sci-fi training montage:

  • Spot takes out the trash: Argon controlled a Boston Dynamics Spot robot dog with an arm. It built a simulated copy of a foot-pedal trash bin, practiced, then stepped on the real pedal and dropped soda cans inside.
  • Mug rescue: Argon guided a simple robot arm to set a fallen mug upright and put a crumpled napkin inside.
  • Sketch to simulation: from a photo of a pen-and-paper sketch, Argon built a physics scene where an Apollo 2 humanoid throws a ball at a stack of boxes.
  • Video labeling: Argon labeled a 136 second video of a two-armed ALOHA 2 robot folding an origami fox, flagging pauses and fumbles.

The practice runs use MuJoCo, a physics simulator. The robot can fail in a virtual world as often as it needs, and nobody has to pick up the cans.

What's next: these are research demos, not a product you can buy. But they show Google wants one model to handle both your emails and a robot's chores.

What it means for you

  • If you pay for AI Pro or Ultra: watch for Argon in the Gemini app and AI Mode. It may take a few days to reach your account.
  • If you use Gemini for free: nothing changes yet. Google hasn't said when Argon reaches free or AI Plus users.
  • If you build apps: test Argon before November 30, while the API price is half of what it will be in December.
  • If you code on a Mac: the Gemini macOS app now includes Antigravity's coding interface, so you can switch from chat to building in one place.

The bottom line

Gemini 4 Argon has moved from a locked room for security experts to the app millions of people use, at least for paying subscribers. Google's main promise for everyday users is simple: shorter, more direct answers. The robot demos hint at where the same model could go next.

Key facts

Who gets it in the app
Google AI Pro and AI Ultra subscribers
API price until November 30
$2 input, $10 output per million tokens
API price after November 30
$4 input, $20 output per million tokens
Robotics benchmark (ERQA)
87%, per Google DeepMind
Apps Gemini connects to
More than 40

Got questions?

Quick answers, plain words

What is Gemini 4 Argon?

It's Google's newest frontier AI model, announced on September 30, 2026. Google built it for long, complex work like software engineering, enterprise knowledge work and cybersecurity defense, and it can write up to 1 million tokens in a single answer.

Is Gemini 4 Argon free?

No. In the Gemini app and AI Mode in Search, it's available to Google AI Pro and AI Ultra subscribers. Android Police notes that free users and AI Plus subscribers don't have it yet.

How much does Google AI Pro cost?

Engadget reports that Google AI Pro costs $20 a month in the US, and the Ultra tier costs at least $100 a month.

How do I use Argon in the Gemini app?

If you subscribe to AI Pro or Ultra, update the Gemini app and look for Argon in the model picker. 9to5Google reports the picker now lists 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro and 4 Argon. The rollout is gradual, so it may take a while to appear.

How much does Argon cost for developers?

In the paid tier of the Gemini API, Argon costs $2 per million input tokens and $10 per million output tokens until November 30, 2026. After that, it costs $4 and $20, according to 9to5Google.

What changed since the September launch?

On September 30, Google released Argon only to vetted cyber defenders through its Fairwind Program. Now it's rolling out to paying consumers in the Gemini app and AI Mode, and to developers through the Gemini API and Google Antigravity.

What is the new tasks toggle?

It's a switch in the Gemini app's main chat window that moves you from chatting to giving Gemini tasks. Google says Gemini can then keep working even when you're not at your computer. 9to5Google reports the feature was previously called Spark.

What does Argon do for robots?

Google DeepMind showed Argon controlling a Boston Dynamics Spot robot and a robotic arm, building physics simulations from camera feeds and sketches, and labeling robot training videos. It scored 87% on the ERQA physical reasoning benchmark.

How does Argon compare to OpenAI's models?

Engadget cites Artificial Analysis, a company that benchmarks AI models, which says Argon matches OpenAI GPT-6 Astra on its Intelligence Index at 60% of the cost per task at introductory prices.

Can I code with Argon in the Gemini app?

Yes, on a Mac. Google is adding the agentic coding interface of Google Antigravity, its AI development platform, directly into the Gemini macOS app. The standalone Antigravity desktop app and command line tool require AI Ultra for Argon.

SourcesGoogle, Google DeepMind
Topics and tagsGoogle, google, gemini, ai models

The daily newsletter

Tech news you'll actually get.

One short email a day. Five minutes. Plain words. Free, every morning.

Free. One email a day. Unsubscribe anytime.

Go deeper

More in brief