Google launches Gemini 4 Argon with $2 per million input tokens introductory price

On October 13, 2026, Google unveiled Gemini 4 Argon, a new model in the Gemini family that the company describes as a frontier model for software engineering, enterprise knowledge work and cybersecurity defense. The model is reaching developers, businesses and paying subscribers at the same time.

Google's post for developers describes Argon as a model for long, multi-step workflows that require deep reasoning and structural precision. It cites code migrations and frontier research as examples.

Where the model is available

Developers and businesses can use Argon through the API in Google AI Studio and the Gemini platform, as well as in Android Studio, on the paid API tier. Google AI Ultra subscribers get it in Google Antigravity, Google's agent-based development platform. Ultra and Pro users will see it in the Gemini app and in AI Mode in Search.

Google states a limit of 1 million tokens combined for input and output, along with a new continuation protocol for long generations. The company also says more models in the Gemini 4 family are on the way.

Pricing and the deadline

The introductory price is $2 per million input tokens and $10 per million output tokens, with cached input tokens 95 percent cheaper. After November 30, 2026, the rates are set to rise to $4 and $20. For companies planning deployments, that means production budgets should already be calculated at the higher price.

What Google claims

According to Google, Argon ranks first on the Vals Index, which covers finance, coding, legal and tax work, and scores 87 percent on the ERQA embodied reasoning benchmark. On the Gray Swan prompt injection test, attack success fell to 0.7 percent across 15 adaptive attempts. These are results cited by the vendor in its launch material.

Google says the Chrome team used the model to separate core webpage handling from a folder of roughly 6 million lines of code, within a Chromium codebase of 40 million lines. The work took weeks instead of months. According to Google, internal teams using Argon in Antigravity increased agentic code submissions by more than 35 percent and cut code rollbacks by nearly a third.

Google says the model is already used by Isomorphic Labs, among others, where it shortens the analysis of PDFs and patents from weeks to hours. Wiz also had early access, and Gamow Labs used Argon in rare disease research.

Safety and science

Under its Frontier Safety Framework, Google deployed a multi-layer defense stack. According to Google, external bio red teaming could unlock less than 1 percent of the biological sub-tasks the company tests for. Google also built an automated misalignment monitor that analyzes an agent's reasoning and actions in real time. The company used it in internal deployments, evaluations and training. A model card has been published.

Google is beginning to expand its Fairwind Program, in which trusted cybersecurity defenders get a version of the model without safeguards against cyber misuse, to hundreds of additional partners. The approach involves an obvious trade-off, so its consequences will only become clear in practice.

For Polish development teams, two things are worth checking: the real cost after November 30, and whether independent tests confirm the claimed coding results.