Gemini 4 Argon goes GA in the API and Copilot

If you have been waiting to try Argon on a real coding project, the availability changed on October 13, 2026. Google made gemini-4-argon generally available, and GitHub added it to Copilot. Choosing it in an editor and migrating an API application are separate jobs. Google’s dated release, GitHub’s announcement.

Check the request you already send

Google’s migration guide says explicit context caching is unsupported. An application that creates caches or passes cached_content needs those calls reviewed. The model also ignores the old sampling fields temperature, top_p and top_k. Thinking can be set to low, medium or high; medium is the default.

For a creator building a small website, use an existing bug as the first comparison. Keep the expected behavior and acceptance check fixed. Then compare the proposed change, whether it passes that check and the billed usage. This is a suggested evaluation, not a task we ran. A model-picker label alone will not tell you whether the new setup helps your project.

The introductory price has an end date

Google’s Standard API prices are in USD per million tokens. Output includes thinking tokens. Checked pricing.

An illustrative request using 100,000 uncached input tokens and 10,000 billed output tokens costs $0.30 at the introductory rates: $0.20 input plus $0.10 output. The same token counts would cost $0.60 at the later rates. This excludes tool charges and is not a measured cost per successful coding task.

Google lists this model as unavailable on the Free tier and Google AI Pro. Do not treat the API price as a consumer-app subscription price.

In Copilot, check access first

GitHub lists Pro+, Max, Business and Enterprise access, with a gradual rollout across supported clients. Business and Enterprise administrators can control the model policy. If the option is missing, check entitlement and policy before troubleshooting your prompt. Availability and controls.

Copilot’s billing reference describes token-based AI credits and separate legacy annual-plan arrangements. Review your own allowance before a long job.

For the earlier model comparisons, see our September Argon benchmark coverage. For this release, start with one bounded task and inspect the configuration and bill before moving a working application.