Gemini 4 Argon: Designed to Tackle Your Toughest Challenges

Global Availability and Pricing

Google DeepMind announced today that its Gemini 4 Argon model is now globally available for developers and enterprises.

Argon can be accessed through the API in AI Studio and the Gemini platform for any developer or business customer.

The model is also offered to Google AI Ultra subscribers via the Antigravity environment.

Google AI Ultra and Pro subscribers can use Argon inside the Gemini app and through AI Mode in Search.

Argon is positioned as a frontier model built for deep work on complex, real‑world tasks.

The model targets high‑impact domains including coding, finance, legal, cybersecurity and scientific research.

In independent benchmark testing Argon ranked number one on the Vals Index, which measures performance across critical economic sectors.

Google introduced an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95%.

Engineering Applications at Google

Google reports that Argon is already helping internal teams solve major engineering problems, streamline business workflows, patch cybersecurity gaps and drive scientific breakthroughs.

One internal use case involved refactoring a large portion of Chrome’s legacy code.

The Chrome team used Argon to separate core webpage handling from a roughly six‑million‑line folder inside Chromium’s forty‑million‑line codebase.

Argon generated multiple clean code changes at once, compressing months of architectural refactor into weeks.

The resulting capability will let billions of Chrome users move pages between tabs, side panels and floating windows without reloading, reducing a class of browser crashes.

Another example is the optimization of a production Python codebase for Google’s open‑source Agent Development Kit (ADK).

The ADK team gave Argon a rulebook and a brief goal to eliminate framework overhead without breaking existing applications.

Working autonomously overnight, Argon produced optimizations now live in the google/adk‑python repository.

Those changes cut in‑memory session preparation time by 40% and sped up cold imports by 21%.

Scientific Accelerations and Model Efficiency

In the scientific arena, Argon accelerated the open‑source quantum chemistry package xTB.

Researchers used Argon in Antigravity to make xTB up to 2.3 times faster at predicting molecular structures.

Argon also achieved a 1.8‑fold speed increase for conformer‑search tasks.

The improvements have been contributed to the official xTB GitHub repository.

Argon further boosted inference speed for Google’s Gemma models.

By running Gemma 4 E2B on an A100 GPU through vLLM, Argon raised throughput from roughly 37,900 to over 108,000 tokens per second.

The speedup represents a 2.85‑fold increase with no loss of token parity or quality, achieved in days rather than months.

These internal results illustrate Argon’s capacity to handle multi‑step, high‑precision tasks across sectors.

The model’s availability through standard APIs means external developers can integrate the same capabilities into their products.

The low introductory cost lowers the barrier for startups and enterprises to experiment with frontier‑level AI.

By offering cached‑token discounts, Google encourages reuse of prompts and iterative development.

The combination of speed, accuracy and pricing positions Argon as a competitive option against other frontier models.

Why This Matters: Argon brings high‑performance, cost‑effective AI to a wide audience, potentially accelerating innovation in software, finance, security and science.