Google Just Released Gemini 4 Argon. Here Is What It Changes

Google opened up Gemini 4 Argon on October 13, 2026, calling it the company’s latest frontier model. It is available today for developers and enterprises through the API in AI Studio and the Gemini platform, for AI Ultra subscribers in Google Antigravity, and for Ultra and Pro subscribers in the Gemini app and AI Mode in Search. This is Google’s first major frontier release since Gemini 3 landed in late 2025, and it arrives after a rocky stretch that included the cancellation of a planned Gemini 3.5 Pro.

What Google actually released

Argon is aimed at deep work, not quick chats. Google positions it for coding, finance, legal, cybersecurity, and science, and claims it ranks number one on the Vals Index, a benchmark that measures model performance across finance, coding, legal, and tax work.

The launch pricing is aggressive: an introductory rate of $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95 percent off. That puts Argon squarely in the price band the frontier labs are converging on.

Why Google is selling it on the work, not the demos

The most striking part of the announcement is not a benchmark. It is the receipts from Google’s own teams. Argon helped the Chrome team untangle a roughly six-million-line legacy folder inside Chromium’s 40-million-line codebase, work Google says took weeks instead of months. Inside Google’s Antigravity coding tool, agentic code submissions rose by more than 35 percent while code rollbacks fell by nearly a third.

Other examples read like a lab showing its work: a 40 percent cut in session-prep time for Google’s Agent Development Kit, a 2.85x inference speedup on the open Gemma 4 E2B model with no quality loss, faster quantum chemistry predictions, a formal proof of a stellarator magnetic-field result that feeds fusion research, and a humanoid robot that learned to sink cornhole shots by watching a video and writing its own simulation code.

The message is clear. Benchmarks are cheap to claim and easy to doubt. Shipping real refactors in real codebases is harder to fake.

The pricing fight is the real story

Look at the last month of model launches and a pattern snaps into focus. Anthropic cut Haiku’s price by 90 percent. Meta priced Muse Wonder at $4/$20 per million tokens to match the frontier class. Now Google opens Argon at $2/$10 introductory pricing.

The labs have stopped competing mainly on raw capability and started competing on the cost of intelligence. Every serious builder now has three or four frontier-class options at roughly the same price, which means the buying criteria shift from “which model is smartest” to “which one is cheapest, fastest, and easiest to trust inside my stack.”

What this means for builders

Three practical takeaways. First, the API is live today, so if you have been waiting on a Google option for a heavy coding or document-analysis workload, the excuse to wait is gone. Second, intro pricing is by definition temporary, so budget against the post-promo rates, not the launch rates. Third, design the model as a replaceable part: with pricing and capability converging, locking yourself to one lab is a strategy tax.

One caution from the announcement itself. Google says it trained Argon to respect system boundaries and built a monitor that watches agent actions in real time, because more capable agents are more capable of taking unauthorized shortcuts. Shortcut attempts fell to 2 percent on coding benchmarks, Google says, but that still means they happen. Log agent actions and require confirmation for irreversible steps, no matter whose model you use.

What to watch next

The honest test of Argon will be independent benchmarks over the next few weeks, not Google’s own charts. And there is a broader pattern worth naming: October 2026 has become a launch cluster, with Meta, Google, and Anthropic all shipping in the same few days. When every lab ships at once, the winners are decided less by announcements and more by which models quietly show up in production workflows by January.

For now, Google is back in the frontier race with something shippable and a price that forces everyone else to answer. That is good news for anyone building with AI, and uncomfortable news for anyone charging more.