Gemini Argon lands in Copilot. Compare the finished work, not just the model.

A new option, with an access check

GitHub announced Gemini 4 Argon in Copilot on October 13. It is rolling out to Pro+, Max, Business and Enterprise users across supported experiences, including the CLI, coding agent and major editors. GitHub says its early evaluations show stronger coding results than previous Gemini models. Those are vendor findings, not our independent tests.

The practical detail is billing: GitHub says the model uses provider list pricing under usage-based billing. Rollout is gradual, and organizational policies can affect access. Check your account’s model picker and current pricing before planning a project around it.

Source: GitHub: Gemini 4 Argon in GitHub Copilot ↗ · Harvard Business Review Press: Power and Prediction ↗

Our take: score the outcome you needed

A model can produce an impressive answer and still leave you with a lot of cleanup. For a small project, the useful comparison is the cost of reaching a result you can accept. That includes your review time, repeated prompts and corrections, as well as whatever the service bills.

Try a task with an obvious finish line: add a sort control to an existing list, preserve its search behavior, and handle an empty result. Write those conditions before asking for code. Afterward, count which conditions were met and record what you had to fix. A long explanation earns no extra credit if the list stops working.

For a broader look at how AI changes decisions and the systems around them, Power and Prediction (affiliate link) is optional background reading. You can start the comparison today with a plain note and the tools you already use.

As an Amazon Associate I earn from qualifying purchases. This optional recommendation is not a claim of hands-on testing.

Keep the comparison fair

Use the same starting project and acceptance criteria for each option. Start from a clean copy so one model does not inherit the other’s repairs. Keep secrets and customer information out of the exercise, and review changes before using them in a real app.

Record the billable usage information your service actually provides; do not guess an exact cost from the length of a reply. If two outputs meet the same requirements, your review effort may be the deciding factor. If neither does, shrink the task and improve the instructions before spending more. This is a proposed comparison method, not a result from testing Argon.