Argon is a noble gas, famously inert, perfectly happy to sit inside a lightbulb and mind its own business. Google's Gemini 4 Argon is billed as the opposite: a model that "works longer and harder" on complicated, multi-step problems. Naming things is hard; shipping them is apparently going a bit better.

From Restricted Preview to Everyone's API Key

On October 13, Google made Gemini 4 Argon broadly available. Developers and enterprises get it through the API in AI Studio and the Gemini platform, Google AI Ultra and Pro subscribers get it in the Gemini app and AI Mode in Search, and Ultra subscribers also get it in Google Antigravity. Google is aiming it squarely at coding, finance, legal, cybersecurity and science work, with a 1M unified input-output token limit via the API.

The benchmark claims are suitably confident: the top spot on the Vals Index (finance, coding, legal and tax), 87% on the ERQA embodied reasoning benchmark, and a 0.7% attack success rate on the Gray Swan indirect prompt injection benchmark. Internally, Google says agentic code submissions rose more than 35% and code rollbacks fell by nearly a third.

The Intro Price Comes With an Expiry Date

Circle this one in red: introductory API pricing is $2 per million input tokens and $10 per million output, with a 95% discount on cached input. On November 30, 2026, that doubles to $4 and $20. If you're prototyping a business workflow on Argon this month, budget for the December invoice, not the launch-week one.

The prompt injection number deserves more attention than the leaderboard bragging. For businesses wiring AI into contact forms, CMS content and support inboxes, indirect prompt injection (malicious instructions hidden inside content the model reads) is the attack that actually keeps security teams up at night. A lower success rate is genuinely welcome, but 0.7% isn't zero, and a vendor's benchmark isn't your production traffic. Run your own evals on your own data.

Shiny new model, same old rules: test it on real work, cost it at full price, and never give it permissions you wouldn't give an unsupervised intern.

Weighing Argon, or any frontier model, for your site or content workflow? We'll help you test it against your actual use case and build the guardrails in before November's price bump turns a pilot into a budget meeting.

Source: Google Blog