Gemini 4 Argon FAQ: every question, answered simply

Gemini 4 Argon is Google's newest frontier AI model and the first model of the Gemini 4 generation. Google DeepMind built it for real-world coding, enterprise knowledge work and cyber defense.

Google DeepMind. It was announced on Google's blog by Koray Kavukcuoglu, SVP at Google DeepMind and Google's Chief AI Architect.

Google hasn't explained the name. Press coverage describes it as a new naming scheme for Gemini models, replacing labels like Pro and Flash for this release.

No. Argon is the first Gemini 4 model Google has announced. Any "Gemini 4 Pro" you read about is speculation.

No. Gemini4Argon.com is an independent guide. It is not affiliated with, endorsed by or sponsored by Google or Google DeepMind. It is published by the team behind Lumeta AI.

It was announced on September 30, 2026. The same day it became available to trusted cyber defenders in Google's Fairwind Program. Wider access is rolling out in stages.

Members of Google's Fairwind Program for cyber defenders, and teams inside Google. Paid API customers and Google AI Ultra subscribers are next.

Google has not announced a free tier. The first consumer access is for Google AI Ultra, a paid plan, and first API access is for paid customers.

Not yet for the public. Google AI Ultra subscribers are among the next to get it, "soon".

It is the program Google uses to give Gemini 4 Argon to trusted cyber defenders first, so they can find and fix security holes before the model is widely available.

Not publicly yet. Google says paid API customers will get access soon.

Google hasn't published one yet, and Argon isn't on the Gemini API models page. Avoid hard-coding a guessed name.

At launch, $2 per million input tokens and $10 per million output tokens, with cached input 95% off. The standard price afterward is $4 input and $20 output per million tokens.

Google confirmed an output limit of 1 million tokens, up from 64K. It has not clearly stated the input context window in its announcement.

The Gemini API has an OpenAI-compatible endpoint, so it should be easy to try from OpenAI-style code once access opens. Google hasn't confirmed Argon on that endpoint yet.

Google hasn't announced Vertex AI timing for Argon yet.

On Google's own table it comes first on 13 of 19 benchmarks, including DeepSWE v1.1 (77.9% vs 74.2% for Claude Opus 5.5 and 74.1% for GPT-6 Astra). It trails on five, mostly terminal and computer-use tests. Independent results aren't out yet.

Long knowledge work (finance, legal and business automation), reading very large inputs, understanding long videos and finding security vulnerabilities, based on Google's published results.

Yes. It takes text, images and video as input, including charts and visual documents. Google reports 91.7% on LVBench, a long-video understanding test.

Google hasn't said it generates images or video. For that, Google's models are Veo 3.1 and Gemini Omni for video and Nano Banana 2 for images, all of which you can use today on Lumeta.

Very, on most of Google's tests: it leads DeepSWE v1.1 and Vibe Code Bench. But Claude Opus 5.5 is well ahead on Terminal-bench 4.0, and GPT-6 Astra leads FrontierSWE v2.

Once you have access, through the Gemini app on Google AI Ultra, or through the Gemini API with an API key from Google AI Studio. Until then, build on the current Gemini model and switch the model name later.

Yes, through a small MCP server that calls the Gemini API. Claude Code and Claude Desktop can then use Gemini as a tool. It works today with current Gemini models and switches to Argon by changing one setting.

Use Google's video models, Veo 3.1 or Gemini Omni. Argon can help you plan the script and shot list, and review footage, but the video itself comes from a video model.