Gemini 4 Argon arrives in Gemini app for Pro and Ultra subs

Gemini 4 Argon is available to Google AI Pro and Ultra subscribers worldwide. In the US, Pro costs $19.99 a month and Ultra costs at least $100. Subscribers can pick Argon in the Gemini app and will also see it powering AI Mode, the chatbot-style answer mode inside Google Search.

Ultra subscribers also get it in Antigravity, Google’s agentic coding tool (software where the AI writes, runs and fixes code with less hand-holding). Developers and business customers can reach the model through the Gemini API and AI Studio.

Free users and Plus subscribers are left out. Google hasn’t said when, or even whether, Argon will come to those tiers. Before the public launch, the only people outside Google using it were members of the company’s Fairwind Program, a group of governments and trusted partners who tested its cybersecurity skills for risks first.

For developers, Argon launched at $2 per million input tokens (the text you send) and $10 per million output tokens (the text it writes back). Those are introductory rates valid until November 30. From December, both figures double.

Even at double, Argon undercuts its main rival on paper. OpenAI’s GPT-6 Astra lists at $10 per million input tokens and $50 per million output tokens. One catch: models use different numbers of tokens for the same job, so a cheaper per-token price doesn’t always mean a cheaper bill. Benchmarking firm Artificial Analysis found Argon matched GPT-6 Astra on its combined Intelligence Index score at about 60 percent of the cost per task, using the launch prices.

For subscribers there is nothing extra to pay. Argon simply becomes part of what Pro and Ultra already include.

Google says Argon gives more direct answers, with less hedging, fewer repeated points and fewer stock phrases, three common complaints about AI chatbots. On long jobs, it shares its progress more often instead of leaving you staring at a spinner, and it can start returning answers while it is still working.

The app itself picked up changes alongside the model:

• a simple switch in the main chat window between normal chat and tasks, where Gemini works on a job for you even when you’re away from your computer;
• Antigravity’s coding interface built into the Gemini app for Mac;
• connections to more than 40 outside apps, such as Shopify, Instacart, Adobe and OpenTable, for things like ordering groceries, booking a table or building a business.

Google claims Argon beats GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 on many tests in coding, cybersecurity, science, finance and law. Those are Google’s own benchmark choices, so treat them as a starting point rather than a verdict.

The company leaned hard on internal use. Google says Argon helped its engineers raise the amount of AI-written code by 35%, find and patch security holes, cut in half the computing power behind YouTube’s comment section, and debug the drone physics simulator used by Wing, Alphabet’s drone delivery company. In a robotics demo, Argon watched a person playing cornhole, worked out the room layout and the player’s movements, and passed that to a humanoid robot that then landed seven throws in a row.

Google DeepMind’s Gemini product lead, Tulsee Doshi, described the model as a “powerhouse” and said Argon is the first of a family of new Gemini models. The launch wasn’t spotless: there were reports of doubts among some Google staff about how well it codes, and the company had earlier dropped its planned Gemini 3.5 Pro model in favor of Argon.

If you already pay for Google AI Pro, just switch to Argon in the model picker and use it. You lose nothing, and the more direct answers and progress updates should help most with longer research, writing and coding jobs.

If you’re a free user, Argon alone probably isn’t a reason to start paying $19.99 a month unless you do a lot of complex work with AI. Everyday questions, recipes and quick summaries won’t feel very different. If you’re comparing paid chatbots, it’s worth a one-month trial against the one you already use, on your own real tasks.

For developers, the timing matters. Any testing you plan to do is cheapest before November 30, after which the bill for the same workload roughly doubles. Check how many tokens Argon actually uses on your jobs, not just the list price.

On safety, Google says it monitors the model’s chain of thought (the step-by-step reasoning it writes out before answering) so it can stop the model if it starts heading somewhere it shouldn’t. That system was part of why the public release waited for the Fairwind tests to finish.