Gemini 4 Argon explained: what Google has said so far

Google has released a new top-tier AI model, Gemini 4 Argon, but you probably can't use it yet. It is going first to a small group of security specialists, with wider access promised "soon".

On 30 September 2026, Google DeepMind announced Gemini 4 Argon. It is the company's first new frontier model generation since Gemini 3. A frontier model is one of the most capable systems a lab has built. Google says Argon is aimed at long-horizon coding (big programming jobs that take many steps), legal and finance work, and cyber defence.

Google is rolling it out in stages:

Argon can now write up to 1 million tokens in one response, up from 64K. Tokens are the small chunks of text a model reads and writes. Google bills this as industry-leading.

Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input (text the model has seen before) 95% cheaper. After the introductory period, the price rises to $4 for input and $20 for output per million tokens.

Google also shared benchmark scores, which are standard tests for comparing models. These are the company's own figures:

Google positions Argon against OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5.

Koray Kavukcuoglu, Google DeepMind's SVP and Chief AI Architect, wrote in the launch post: "Argon is helping our quantum computing researchers optimize the spacetime resources (qubits × gates)"

If you use Gemini in the consumer app, nothing is confirmed for you yet. Google has given no date for the consumer release.

If you pay for Google AI Ultra or use the paid API, you are in the second wave of access. If you build software, the pricing matters. The $2 and $10 rates are introductory, and Google has said they will double afterwards.

Treat the benchmark results with care. They come from Google, not independent testers. CNBC asked whether Argon can catch the frontier. One analyst told the outlet that "the final proof will be in enterprise production environments once the model is fully released". Bloomberg also reported that some Google employees privately doubt its real-world coding ability. Google has not confirmed that report.

Google says broader developer, enterprise and consumer access is coming, but it has not given dates. Real-world results from businesses using Argon after full release will show how it compares with GPT-6 Astra and Claude Opus 5.5.