Google launches Gemini 4 Argon, which it calls its latest frontier model
Google says the model outscores other frontier models in benchmark tests spanning different domains; at its introductory price, input costs $2 per million tokens and output costs $10 per million tokens.
AI illustrationGoogle has launched Gemini 4 Argon, which it calls its latest frontier model, and says it outscores other frontier models in benchmark tests spanning different domains. According to the company, Argon sits at the top of the Vals Index, which tracks model performance in economic sectors such as finance, coding, law and tax.
The model became available globally on 13 October. Developers and enterprises get API access through AI Studio and the Gemini platform. Subscribers to Google AI Ultra and Pro can use it in AI Mode in Search and the Gemini app, while Ultra subscribers also get it in Google Antigravity. At the introductory price, input costs $2 per million tokens and output costs $10 per million tokens, while cached input tokens are 95% cheaper.
Google said its teams using Argon in Antigravity saw agentic code submissions rise more than 35%, while code rollbacks fell by nearly a third. In Chromium's codebase of 40 million lines, the Chrome team used Argon to pull core webpage handling out of a legacy folder of about 6 million lines, compressing months of work into weeks. Google said the change will allow live webpages to shift among tabs, floating windows and side panels without reloading.
Google researchers used Argon in Antigravity to speed up xTB, a much-cited open-source package for quantum chemistry. Its molecular structure prediction became up to 2.3x faster and its conformer search 1.8x faster, the company said. Google said it also used Argon to make inference on its open Gemma 4 E2B model 2.85x faster end to end on an A100, lifting throughput from about 37,900 tokens per second to over 108,000. It said the work took a few days, whereas it would usually need months.
On security, Google said it has begun expanding to hundreds more Fairwind Program partners as it brings advanced cyber capabilities to a wider circle of trusted defenders. It said Wiz, which got early access to Argon, used the model to protect healthcare infrastructure, where it helped uncover and swiftly close a public file download bypass; the flaw had left a master key exposed and allowed administrator access.
Google also said Gamow Labs researchers used Argon to study the de-identified genomes of newborns who have rare developmental lung disorders. In one case, it said, the model found a chromosomal rearrangement that had escaped routine software for years.