Google’s Gemini 4 Argon is now available to paying Gemini subscribers, bringing the company’s new flagship AI model to consumer apps as well as professional development tools. Released on 13 October 2026, Argon is designed for work that takes time and several steps to complete, from investigating a software problem to analysing a collection of research documents.
For UK readers, there is an important distinction between using Argon in the Gemini app and building software with it. Google AI Pro and Ultra subscribers can access the model through Gemini and AI Mode in Search. Developers have a separate, paid route through Google’s API, while Ultra subscribers also receive access to Argon in the Antigravity coding environment.
Where UK subscribers can access Gemini 4 Argon
Google says its Gemini 4 Argon rollout began globally on 13 October for Google AI Pro and Ultra customers. That covers the Gemini app and AI Mode in Google Search, although individual features may still depend on the account, product and rollout stage. Anyone struggling to find Argon should check that they are signed into the correct Google account with an eligible subscription.
Google AI Ultra adds access to Google Antigravity, an agent-based development environment intended to help with coding projects. Antigravity’s coding interface has also been integrated into the Gemini app on macOS. It lets supported users move from discussing a project to asking a coding agent to work on it, with the agent’s proposed actions available for review. That is not a promise that the same tools are present on every phone or subscription tier.
The other route is aimed at software developers. Google’s Gemini API documentation lists the model as gemini-4-argon, available through its paid API tier. Access through a consumer Google AI Pro subscription does not mean free or bundled API usage. Google AI Ultra customers can also try Argon in AI Studio, but anyone building a service around the model should check the separate API terms and charges.
What can the new model do?
Argon’s main selling point is its ability to follow a difficult problem for longer. Google is targeting software engineering, mathematical and scientific work, professional document analysis and tasks involving several connected tools. The idea is to get beyond a polished answer to a prompt and produce something usable: a revised piece of code, a structured analysis or a completed multi-stage job.
Google has given some concrete examples in its launch announcement. Its Chrome engineers used Argon to help separate legacy browser code into more manageable components. Another team applied it to the open-source Agent Development Kit, reporting faster imports and lower overhead during lengthy sessions. Those are Google’s own results from projects it knows well, not guarantees that every developer will see comparable gains.
There are more familiar uses too. Google describes turning scattered project material into a proposal presentation, examining dense reports and working across connected apps. A user could, for instance, ask for information from several documents to be organised into a useful first draft, then check the citations, figures and conclusions before sharing it. For most people, that sort of time saving will matter more than a leaderboard position.
The Gemini app is also gaining a clearer split between ordinary chat and tasks. In the latter, Gemini can work with connected services to carry out approved steps. Google says it supports connections to more than 40 apps, although the exact integrations available will vary. Before handing it a complicated job, users should check which accounts are linked and what they have authorised Gemini to do.
Microsoft is pursuing a related idea through its hybrid AI plans for Windows, which combine processing on a PC with cloud services. The products work differently, but both companies are trying to make AI useful inside everyday workflows rather than confining it to a separate chat window.
Why developers are interested in the one-million-token limit
Google’s published specification gives Argon a shared context window of 1,048,576 tokens. It can process text, images, audio, video and PDF input through supported API features. That amount of context creates room for a large codebase or lengthy source material to be examined together, reducing the need to split every part of a project into disconnected requests.
There is a catch behind the headline figure. Input, the model’s reasoning and generated output all share that allowance. A normal synchronous API request has a maximum output of 65,536 tokens. Longer responses require Google’s background or continuation options. It would be misleading to suggest that a consumer can simply ask the Gemini app for a million-token answer.
Capacity is not accuracy, either. Google DeepMind acknowledges in its model card that Argon can still make factual mistakes, while complex requests can be slow or time out. Large documents may fit inside the context window, but important claims and calculations still need checking against the original material.
How much does Gemini 4 Argon cost in the UK?
Consumers need an eligible Google AI Pro or Ultra subscription. Argon is a feature of those plans, not a separately priced monthly product. Google provides the current tiers and offers on its UK Google AI plans page. Pricing may differ from US promotions, and a reliable live UK checkout price was not established for this report, so we are not quoting an unverified pound figure.
For developers, Google has published an introductory API rate of US$2 per million input tokens and US$10 per million output tokens, with reasoning tokens included in the output calculation. Those prices apply through 30 November 2026. From 1 December 2026, Google’s stated standard rates are US$4 and US$20 respectively. The developer guide also explains discounted cached input and the technical changes needed to use the model.
A short API experiment may cost little, but regular requests containing substantial documents or producing long outputs can accumulate charges. Anyone planning to deploy Argon in an application should estimate likely token usage before relying on the introductory rate. Google AI Pro and Ultra subscription fees are separate from those metered developer charges.
Google’s benchmarks look strong, but there is no clean winner
Google reports a 68.9% score on the Vals Index, ahead of the rival models included in the table published with Argon’s model card. That benchmark is intended to assess professional knowledge work, including legal, financial and coding tasks. The same table also contains tests where competing models score higher, particularly in some software engineering and scientific evaluations.
There is a difference between leading a particular test and being better at everything. Results depend on the task, evaluation method, available tools and prompting. They also say relatively little about a reader’s everyday priorities, such as response speed, dependable source handling and subscription value. Independent comparisons will be more useful now that a wider audience can put Argon through its paces.
What about reliability and security?
Google has outlined additional safeguards for Argon, including measures intended to resist prompt injection, restrict misuse and monitor unexpected agent behaviour. Its safety announcement also describes work with specialist cybersecurity defenders. These are important parts of the release, but they do not remove the risks of giving an automated tool access to files, messages or business systems.
The most useful starting point is a task with an outcome you can judge. Ask Argon to compare two documents you know, investigate a reproducible software issue or organise an existing project. Check the result and its supporting material before allowing any connected tool to send messages, change important files or act without further review.
Tech Torque Verdict
Google has made Argon available in places people already use, and that is more persuasive than a dramatic benchmark headline. If the model can reliably help someone work through a complex document, fix a stubborn coding problem or finish a multi-stage task, it has a practical reason to exist beyond the AI race.
The access rules are less straightforward. Pro and Ultra subscribers can try the model in Gemini, but developer API usage has its own paid tier, and the more advanced coding environment is an Ultra benefit. Anyone considering a subscription specifically for Argon should check the features available to their account before paying.
Argon is worth testing, especially for subscribers who already use Gemini regularly. Google has published encouraging results, but the case for switching tools or spending more money will be decided by consistent everyday performance, not the launch-day figures alone.
