Google's Gemini 4 Argon Wants to Be the Model Lawyers and Bankers Actually Trust
The model launches at $2 per million input tokens, ranks first on a key industry benchmark, and signals where Google thinks the real AI battle is being fought.
Key points
- Gemini 4 Argon, Google's frontier AI model, launched globally on 15 October 2026 for developers, enterprises, and professional subscribers.
- Pricing is set at $2 per million input tokens and $10 per million output tokens, with a discount on cached inputs for enterprise buyers.
- The model ranks first on the Vals Index, a benchmark measuring AI performance across finance, coding, and legal work.
- Google AI Ultra subscribers can access it through Google Antigravity; AI Pro subscribers get it inside the Gemini app and AI Mode in Search.
This isn't a flashier chatbot. Gemini 4 Argon is built for multi-step, high-stakes professional work: contract review, tax analysis, cybersecurity assessments, complex code. Google calls it "deep work", and that framing is deliberate. We first covered Argon on 2 October 2026, when it was still rolling out; today's pricing and benchmark news are the clearest statement yet of what Google intends it to do.
What does Argon actually do better?
It tops the Vals Index, a benchmark that scores AI models on performance across finance, coding, and legal sectors. Those are domains where errors carry real cost, which is exactly why Google chose that leaderboard to lead with.
Google's positioning is precise: Argon isn't a general-purpose assistant. It's the model a law firm or a software team reaches for when the task is genuinely complicated and getting it wrong isn't an option. Whether professional buyers agree is still an open question, but the intent is unmistakable.
For regular users, the main access point is the Gemini app. AI Pro subscribers get Argon; Ultra subscribers also gain access through Google Antigravity, Google's newer workspace product. Developers can connect via the API, the technical channel that lets software products plug into AI models directly.
How does the price compare?
The introductory rate of $2 per million input tokens and $10 per million output tokens puts Argon in the competitive zone for frontier models.
| Tier | Input price | Output price | Cached input |
|---|---|---|---|
| Argon (introductory) | $2 / 1M tokens | $10 / 1M tokens | Discounted |
A token is roughly three-quarters of a word, so a million tokens covers an enormous volume of text. Most individual users won't touch the API rate at all; the Gemini app subscription is their relevant price point.
Enterprise buyers will pay close attention to the cached-input discount. Cached inputs are chunks of text the model has already processed and stored, so re-using them repeatedly, say a long legal document referenced across hundreds of queries, costs a fraction of the standard rate.
What has changed since our last Google coverage?
Our 13 October story on Argon's API launch noted the model's promise for engineering tasks and business workflows. Today's pricing announcement sharpens that picture considerably: Google now has a number to put next to the ambition.
Argon sits in a distinct lane from the consumer or scientific AI pushes we've seen lately. Competition from OpenAI and Anthropic in professional workflows is already fierce, and enterprise procurement cycles are slow. The real question isn't whether Argon leads the Vals Index; it's whether a benchmark score moves budget decisions inside a law firm or a bank. Google is betting it will. That bet is worth watching.
