Gemini 4 Argon opens its API at launch pricing

Anyone who wants to try Gemini 4 Argon through its API pays half the later rate until Nov. 30. According to Google's announcement, published Oct. 13, the launch price is $2 for every million tokens the model reads and $10 for every million it writes. After that it rises to $4 and $20.

A token is the unit a language model uses to count text, and APIs bill by the million. Google also offers a 95% discount on input tokens that are cached, meaning repeated across requests.

Argon is the most advanced model in the Gemini 4 family. It was unveiled on Sept. 30, but only a group of cybersecurity defenders could use it. Since Oct. 13 it is available in Google AI Studio and on Google Cloud under the identifier `gemini-4-argon`. Google AI Ultra and Pro subscribers can use it inside the Gemini app and in AI Mode in Search. Antigravity, Google's development platform, is for Ultra only.

What the tests show

The ERQA (87%) and LVBench (91.7%) figures come from Google, which sells the model, and no independent replication was found. The results that do come from third parties do not fully agree:

• Vals AI ranks it first on its index, with 68.90% and a cost of $15.68 per test.
• Artificial Analysis gives it 53 points and eighth place among 216 models. It also generated 110 million output tokens in its evaluation, against a median of 85 million.

That last figure matters for the bill. A model that writes more to solve the same task costs more even if its per-token rate is the same, so the list price does not say what a task will cost.

What is still unknown

Google says Argon is available "globally" but does not name countries. For Mexico there is no official confirmation of the API or of the Pro and Ultra plans, and no local price for those subscriptions.

Google's blog ends the launch rate on Nov. 30; the Google Cloud blog speaks of the new rate "from Dec. 1." Check the official price table before budgeting.