Since October 13, 2026, Gemini 4 Argon, Google’s newest frontier model, has reached general availability inside GitHub Copilot. The model targets complex coding work, agentic workflows, and multi-step reasoning, with the clearest gains showing up in long Copilot CLI sessions. Its arrival confirms Copilot’s multi-model direction, where Claude, GPT, and Gemini now sit side by side in the same picker, alongside agents wired together through the MCP protocol.
A model built for long agentic runs
Gemini 4 Argon outperforms earlier Gemini generations across GitHub’s internal coding benchmarks, with the widest margin on long Copilot CLI sessions, where an agent chains several shell commands without constant supervision. The model checks part of its own work: it tests edge cases before marking a task finished, a behavior GitHub compares to the self-review habits already seen in other reasoning models.
Availability across environments and plans
The rollout is gradual and does not reach every account at the same pace. It is nonetheless already selectable across most of the surfaces where Copilot shows up.
| Environment | Model selection | Plans covered |
|---|---|---|
| VS Code / Visual Studio | Built-in model picker | Pro+, Max, Business, Enterprise |
| Copilot CLI | Largest measured gains | Pro+, Max, Business, Enterprise |
| JetBrains / Xcode / Eclipse | Built-in model picker | Pro+, Max, Business, Enterprise |
| Copilot app / coding agent | Gradual rollout | Pro+, Max, Business, Enterprise |
Standard Copilot Pro subscriptions are not mentioned in the announcement, which for now reserves the model for the higher-tier plans. Usage is billed at provider list pricing under usage-based billing, putting Gemini 4 Argon in the same pricing bracket as the other premium models already in the picker.
What administrators can adjust
For Business and Enterprise organizations, access runs through Copilot’s model policy. By default, a new model turns on automatically unless an administrator has disabled default model enablement or blocked this specific model. Inside a Copilot CLI terminal, the choice happens when starting a session:
$ gh copilot
> /model
1. GPT-5.2 Codex
2. Claude Opus 5.5
3. Gemini 4 Argon (new)
4. ...
Selected model: Gemini 4 Argon
In GitHub’s own internal testing, Gemini 4 Argon solves more tasks than earlier Gemini generations on every coding benchmark, with its largest gains on long Copilot CLI sessions.
Gradual rollout means an eligible account (Pro+, Max, Business, Enterprise) may not see Gemini 4 Argon appear in its picker right away. Teams running command-line tools as part of their development workflow should confirm availability before building it into a documented process.
Key takeaways
Gemini 4 Argon joins a Copilot picker already shared by several model families, with a stated edge on long agentic command-line sessions. Availability still sits behind Pro+, Max, Business, and Enterprise plans, billed at provider pricing. Teams experimenting with agents in production are better off testing the model on a representative workflow before rolling it out broadly, since the gradual rollout can make early results vary from one account to another.
The shift to a multi-model picker in Copilot changes the question teams are actually asking: it’s no longer “which AI tool to adopt,” but “which model for which task, at what cost.” I’ve watched several clients make that call model by model depending on the work at hand — refactoring, test generation, exploring an unfamiliar codebase — rather than locking in a single default, and that habit is the one that keeps paying off over time. — Simon Janvier
Primary source: GitHub Changelog, “Gemini 4 Argon in GitHub Copilot,” October 13, 2026.
