On November 18, 2026, GitHub announced the general availability of Gemini 4 Neon in GitHub Copilot, a new Google model positioned for agentic coding, multistep tool use, and knowledge work. It is the fourth Gemini 4 variant to land in Copilot in just over a month, following Argon, Carbon, and Hydrogen.
A cheaper alternative to Gemini 4 Carbon
Per the details Google published and GitHub relayed, Neon targets quality close to Gemini 4 Carbon, the security-focused model that arrived in early November, but at a lower token cost per task and a shorter time to first response. GitHub also highlights fewer unrequested edits to existing code, and less tendency to wrongly assert things about the code, the tools invoked, or test results. These comparisons rest on the provider’s own claims; GitHub has not published independent benchmarks to back them.
Four variants in a month, four different jobs
The release pace has already built out a sizeable model family, each one aimed at a distinct slot in Copilot’s picker.
| Model | General availability | Stated positioning |
|---|---|---|
| Gemini 4 Argon | October 13, 2026 | Complex coding, agentic work, and reasoning |
| Gemini 4 Carbon | Early November 2026 | Secure coding, security-focused generation |
| Gemini 4 Hydrogen | November 13, 2026 | Fast Flash model, replaces Gemini 3.8 Flash (retiring December 14) |
| Gemini 4 Neon | November 18, 2026 | Agentic coding and tooling at lower cost, a cheaper alternative to Carbon |
Four Gemini 4 variants have landed in Copilot’s model picker in just over a month, each aimed at a different trade-off between quality, speed, and cost.
Availability and default activation
Gemini 4 Neon is available on Copilot Pro+, Max, Business, and Enterprise plans, across Visual Studio Code, Visual Studio, Copilot CLI, Copilot’s coding agent, the GitHub Copilot app, github.com, GitHub Mobile, and JetBrains IDEs, Xcode, and Eclipse. The rollout is staged across surfaces. Billing follows the provider’s list pricing as part of usage-based billing.
Pinning a default model at the project level is still possible through the extension’s settings, for example to force Neon on agent tasks while keeping Carbon for security-focused reviews:
{
"github.copilot.chat.agent.model": "gemini-4-neon",
"github.copilot.chat.reviewSelection.model": "gemini-4-carbon"
}
The release cadence mostly reflects the orchestration race Copilot has been running since the MCP ecosystem became the standard way agents pick tools: the more reliably a model chains tool calls, the more cost per task becomes as much a selection criterion as raw quality.
Key takeaways
Gemini 4 Neon rounds out a family of four Gemini 4 models shipped in Copilot since mid-October, explicitly positioned as a cheaper option than Carbon for everyday agentic work. The default activation in Business and Enterprise model policies is worth a quick check for teams watching their AI spend closely.
The growing pile of Gemini 4 variants in Copilot is starting to look like what Claude and GPT already went through: a model picker that gets unreadable for anyone not tracking every changelog entry. The right move is still to pin the model explicitly per task in the project settings, rather than let default activation decide for the team — Simon Janvier.
Further reading: the official announcement of Gemini 4 Neon in GitHub Copilot on the GitHub changelog.
