Google releases Gemini 4 Argon worldwide
Google has launched Gemini 4 Argon, a new frontier AI model that it says is designed for difficult, multi-step work in fields including software engineering, finance, law, cybersecurity and science. The company said the model is globally available from October 13 through its API offerings, AI Studio, the Gemini platform, the Gemini app, AI Mode in Search and Google Antigravity for eligible subscribers.
The release is notable less for a single consumer-facing feature than for the breadth of work Google says it wants the model to handle. Argon is being positioned as a system for “deep work”: tasks that require a model to retain context, reason through several stages and produce useful output in settings where mistakes can be expensive or time-consuming to correct.
Google said Argon ranked first on the Vals Index, a benchmark it described as measuring performance across economically significant domains such as finance, coding, legal work and tax. As with any benchmark claim, the result is best read as one indicator of model capability rather than a guarantee of performance on a particular organization’s data, workflows or risk requirements.
A push toward larger engineering tasks
One of Google’s central examples is software maintenance at scale. The company said its engineers have used Argon on difficult internal projects, including work on Chromium, the open-source browser codebase that underpins Chrome and other browsers.
According to Google, Argon helped the Chrome team separate core webpage-handling functions from a legacy browser-features folder containing roughly 6 million lines of code within Chromium’s approximately 40 million-line codebase. The work involved an architectural refactor across hundreds of files. Google said the model enabled the team to complete work that would otherwise have taken months in a matter of weeks.
The company said the refactor is intended to support webpages moving between tabs, side panels and floating windows without reloading. That example illustrates the kind of AI-assisted engineering Google is emphasizing: not merely generating a standalone code snippet, but helping teams navigate, alter and validate changes across a large, interdependent codebase.
Google also reported that engineers using Argon in Antigravity increased agentic code submissions by more than 35% while code rollbacks fell by nearly a third. Those figures are internal company claims and do not establish that every development team will see the same results. They do, however, signal how major technology companies are increasingly measuring AI tools: by their effect on delivery speed and the rate at which changes need to be undone.
Availability and pricing
Developers and enterprises can access Gemini 4 Argon through Google’s API and platform products. Google said the model is also available to Google AI Ultra subscribers in Antigravity, while Google AI Ultra and Pro subscribers can use it in the Gemini app and AI Mode in Search.
The model launches with introductory pricing of $2 per million input tokens and $10 per million output tokens. Google said cached input tokens receive a 95% discount. Token pricing matters because complex agentic workloads can repeatedly send long documents, codebases or other context to a model; caching can reduce the cost of reusing information across a workflow.
For organizations, availability across developer and consumer surfaces means Argon can potentially appear in several parts of the same technology stack: in a custom application through an API, in a workplace-oriented platform and in a general-purpose assistant. That reach could make adoption easier, but it also raises familiar governance questions about where sensitive information is processed, which tasks can be automated and what human review remains necessary.
Claims across science and business
Google said teams inside the company, enterprise users and researchers have already used Argon for code refactoring, workflow optimization, cybersecurity vulnerability work, quantum computing, nuclear fusion and life-sciences research. The company did not present these uses as a substitute for domain expertise. Rather, it framed the model as an accelerator for specialists confronting complex information and technical processes.
That framing is increasingly important in the competition among frontier-model developers. The focus is shifting from a chatbot’s ability to provide a fluent answer to its ability to contribute reliably to a longer chain of work. In practice, that may mean reading a large repository, preparing a first pass at a legal or financial workflow, identifying potential security issues or helping researchers organize and test ideas.
Google’s announcement also places particular weight on safety and reliability. The company said it developed Argon responsibly and safely, while its examples stress precision in high-consequence domains. The practical test will be how customers evaluate outputs in their own environments, especially where an answer must be checked against current rules, source material, code or experimental evidence.
What the launch changes
Gemini 4 Argon expands Google’s effort to make advanced AI a working tool for developers, enterprises and researchers rather than only a consumer assistant. Its broad release and usage-based pricing give organizations a direct route to test that proposition.
The immediate question is not whether a model can attempt complex professional work. It is whether it can do so consistently enough to fit into real teams’ review processes. Google’s Chromium example suggests the company sees large-scale code maintenance as an early proving ground. If the reported gains translate beyond Google’s own engineering organization, Argon could become part of a wider move toward AI systems that assist with substantial, multi-step technical work.
This article is based on reporting by Google AI Blog. Read the original article.
Originally published on blog.google
