Google has released Gemini 4 Argon, a model aimed at complex coding and other tasks that take many steps to finish. Builders can access it through Google’s tools, GitHub Copilot and Vercel AI Gateway, which can connect it to existing coding agents.
The practical choice is where to use it and how usage gets billed. Access through a subscription and paying for each request are different arrangements.
Working across a codebase
Google reports using Argon to restructure Chrome code across hundreds of files and improve performance in its open-source Agent Development Kit. These are examples of changing how an existing project works while preserving its behavior, the kind of task that gets harder as an app grows.
Google’s reported benchmark results from its September 30 announcement. Open the table to read it at full size.
The coding results vary by test. Google’s table puts Argon ahead on DeepSWE v1.1, but other models score higher on FrontierSWE v2 and Terminal-bench 4.0. These benchmarks measure performance on different sets of tasks, so the table does not establish one model as the best choice for every project.
The useful question for a builder is whether an agent can carry a change through related files, checks and fixes without losing track of the goal. Google’s examples show work inside its own engineering teams; they do not establish how reliably Argon will handle your app with less supervision.
Direct usage prices double after November 30
Google’s Gemini API, the service an app or coding tool uses to send requests to the model, launches at $2 per million input tokens and $10 per million output tokens. Tokens are the units used to measure material sent to the model and generated in response. After November 30, 2026, those rates rise to $4 and $20 respectively.
A coding agent can make repeated requests as it reads files, edits code and runs checks, so one instruction can generate several billable calls. Check the rates in the service you use before budgeting a long task, including the rate that will apply after the introductory period.
Access through Google, Copilot or your existing agent
Google lists access through the paid Gemini API in AI Studio and Android Studio, plus its Antigravity coding tool for Google AI Ultra subscribers. Google AI Pro and Ultra subscribers also get access in the Gemini app and AI Mode in Search.
GitHub’s rollout covers Copilot Pro+, Max, Business and Enterprise. Supported tools include VS Code, Copilot’s terminal tool and the Copilot app. Availability is gradual, so the model may not appear in your picker yet. GitHub bills Argon at provider list prices under usage-based billing.
Business and Enterprise administrators should check model policies if they require approval before use. GitHub enables new models automatically unless an administrator has turned off that default or disabled the model.
Vercel AI Gateway offers another route. It handles model requests and billing, with setup for coding agents including Claude Code, Codex and Cursor. Using Argon this way requires connecting your agent to AI Gateway, then selecting google/gemini-4-argon. Vercel says it passes through provider pricing without a markup or inference platform fee.
Vercel lists a 1 million token context window, the amount of material the model can work with, and text, image, PDF and video input. That allows more project material in a request, but does not guarantee the agent will use every detail correctly.
If you use Gateway to manage spending, its budget limits block later requests after a threshold is crossed. The request that crosses the threshold still completes, so the final bill can exceed the limit.
End of article