Gemini 4 Argon Is Now Generally Available

The short version

Gemini 4 Argon is Google's first Gemini 4-generation model. When it was first announced on September 30, 2026, access was limited to a vetted group of cyber defenders through the Fairwind Program. That changed on October 13: Google marked the model generally available in the Gemini API and Google AI Studio, and separately confirmed it's rolling out to Google AI Pro and Ultra subscribers in the Gemini app^[1][4].

The headline change is scale. Gemini 4 Argon supports a unified 1,048,576-token context window — input, thinking, and output share a single pool — and long decode, meaning a single response can run well past the old 64K-token cap, up to 1M tokens minus your input^[2].

Pricing: $2/$10 intro, then $4/$20

Gemini 4 Argon launches with an introductory price that runs through November 30, 2026, then roughly doubles on December 1^[2]:

Gemini 4 Argon API pricing, October 2026^[2].

Cached input tokens are discounted 95% through implicit caching, which is enabled by default for prompts of at least 4,096 tokens^[2]. One thing to note: Gemini 4 Argon is not available on the Free tier or the Google AI Pro plan — it's aimed at paid API customers and Google AI Ultra subscribers^[2][4].

What changed under the hood

Beyond availability, GA brings several concrete API capabilities that weren't in the earlier preview^[2]:

• Long decode. A single response can now exceed 64K tokens — up to 1M tokens minus your input — using background execution.
• Tunable thinking levels. Pick low, medium (default), or high to trade reasoning depth against latency and token spend.
• Agentic video understanding is GA. Video understanding graduated to general availability alongside the model.
• Managed Agents support. The Antigravity agent in Gemini Managed Agents can now run on Gemini 4 Argon.

If you're migrating from an older Gemini

Google's migration guidance flags a few breaking changes worth knowing before you swap model strings^[2]:

• Raise max_output_tokens. Argon can spend more of the limit on thinking, so Google recommends at least 32,768 tokens.
• Drop sampling parameters. temperature, top_p, and top_k are ignored if you still pass them.
• Remove explicit context caching. cachedContents / caches.create aren't supported; implicit caching replaces them.
• Use background execution for long outputs. Anything past 64K tokens needs background: true.

Bottom line

Verdict: Gemini 4 Argon's GA is the moment Google's frontier model stops being a security experiment and starts being something developers can actually build on. The $2/$10 introductory pricing is competitive for a frontier model — just budget for the step-up to $4/$20 on December 1. If you're not on a paid plan, this still isn't for you; it stays gated to paid API customers and Google AI Ultra subscribers.

Frequently asked questions

When did Gemini 4 Argon become generally available?

October 13, 2026, in the Gemini API and Google AI Studio^[1].

How much does Gemini 4 Argon cost?

Through November 30, 2026: $2 per million input tokens and $10 per million output tokens. From December 1, that rises to $4 and $20^[2].

What is the context window?

A unified 1,048,576-token window shared by input, thinking, and output, with long decode supporting responses up to 1M tokens^[2].

Is Gemini 4 Argon available on the free tier?

No — it's not available on the Free tier or the Google AI Pro plan^[2].