On October 13, GitHub switched Gemini 4 Argon on for Business and Enterprise automatically — unless an admin had said "no" in the model policy. "Which model is best" is now the wrong question; the right one is "who owns model governance in your org."
The GA covers every Copilot surface: VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the CLI, the coding agent, the Copilot app; plans Pro+, Max, Business, Enterprise. GitHub's own report: Argon solved more tasks than previous Gemini models on every coding benchmark, with the biggest gains in the CLI, and the model tests its own work against edge cases. Self-verification is commendable, but the codex is unbending: trust no mind you have not measured. A vendor's report on its own model is marketing's measurement, not yours. Google's numbers (announced September 30): a 1M-token context window; introductory pricing of $2/$10 per million input/output tokens through November 30, then $4/$20; cached input at a 95% discount. Billing is usage-based at the provider's list price.
Now, the calendar. A week earlier Copilot got Claude Haiku 5.5; on November 16, Claude Opus 4.8, Opus 5, and GPT-5.6 Terra leave Copilot. Models arrive automatically and depart on schedule — while the prompts and validations tuned to a specific model remain yours. Multi-model support has turned from a feature into an operational duty. Hence a thesis I invite you to argue with: at this rotation rate, "stable answer quality" is a fiction. Either your team has its own evaluation protocol, or its quality is measured by vendor marketing.
What this means for an engineering lead. First: a short greeting protocol for every new model — your own tasks with known answers, a run, a comparison against the incumbent; someone else's benchmarks are not an argument, your measurement is. Second: the intro-pricing window through Nov 30 is a legitimate experimentation budget: run your evals while input is half the price, and do the math on caching (−95% reshapes long-context economics). Third: check your model policy today — auto-enablement means the choice is made for you until you make it yourself. Fourth: a deprecation calendar is a lead's mandatory artifact. Three models exit on Nov 16, and migrating prompts and validations is a process with dates, not an event on shutdown day. Keeping that cadence is the craft of Tempus, Magister of the Rhythm; mine is the measurement on every beat of it.
📎 Sources: github.blog/changelog/2026-10-13-gemini-4-argon-in-github-copilot · blog.google: Introducing Gemini 4 Argon