A developer studies colorful code and AI-assisted workflow tools on a large monitor in a neon-lit workspace.
Gemini 4 Neon is now generally available in GitHub Copilot. It's the latest Google model to land in the Copilot model picker. GitHub's changelog says it is designed for agentic coding, multistep tool use, and knowledge work. If you run Copilot for yourself or for a team, the useful questions are who can use it, where it appears, how it's billed, and how to switch it off. Most of the performance claims come from Google rather than from independent testing.

A developer studies colorful code and AI-assisted workflow tools on a large monitor in a neon-lit workspace. What GitHub announced​

GitHub framed Neon as a cheaper, quicker sibling to Gemini 4 Carbon. All of the comparisons are attributed to Google:

  • Neon delivers quality close to Gemini 4 Carbon at a lower token cost per task.
  • Compared with Carbon, it starts responding faster and makes fewer unrequested changes on coding tasks.
  • It is also less likely to make unsupported claims about code, tool, or test results.

The changelog gives no benchmark numbers, test methodology or measured savings. A trade publication also noted that the announcement does not quantify the claimed per-task savings. Treat these as vendor claims until you've tried the model on your own repositories.

The "fewer unsupported claims" line is the one agent users should care about most. Anyone who has watched an AI agent report "all tests pass" without running them knows why. Whether Neon actually behaves better here is something to check on your own workloads, not something this announcement establishes.

Who gets it, and where​

Neon is available to Copilot Pro+, Max, Business, and Enterprise users. Copilot Free, Student and Pro aren't named in the announcement. That differs from Gemini 4 Hydrogen, a Flash model announced on November 13, which is available to Copilot Pro, Pro+, Max, Business, and Enterprise users.

GitHub's docs add that Copilot Student and Copilot Free users have access to models through auto model selection only.

You can pick Neon from the model picker in these places:

  • Visual Studio Code and Visual Studio
  • Copilot CLI
  • GitHub Copilot coding agent and the GitHub Copilot app
  • github.com
  • GitHub Mobile on iOS and Android
  • JetBrains IDEs, Xcode and Eclipse

Rollout is gradual, and GitHub's advice is simply to check back if Neon isn't there yet. If you're on an eligible plan and don't see it, work through this list:

  1. Confirm your plan is Pro+, Max, Business or Enterprise.
  2. Update your IDE or extension. GitHub's supported-models page lists minimum IDE versions for recent models, though the entries I saw for the newer Gemini 4 models were still marked "TBD".
  3. Wait for the rollout to reach your account.
  4. On Business or Enterprise, ask an administrator whether the model policy allows it.

Steps 2 to 4 are my own practical checklist, not a procedure from GitHub.

Cost: what the announcement does and doesn't say​

GitHub says Neon is billed at provider list pricing under usage-based billing. The changelog doesn't give the rate, but GitHub's pricing table does. Neon is listed as GA, in the Versatile category, at $3.00 per million input tokens, $0.15 per million cached input tokens and $15.00 per million output tokens.

Here is how that compares with the other Gemini 4 models in the same table:

ModelInputCached inputOutput
Gemini 4 Hydrogen$1.50$0.075$7.50
Gemini 4 Neon$3.00$0.15$15.00
Gemini 4 Carbon (promotional, until Dec 1, 2026)$2.00$0.10$10.00
Gemini 4 Carbon (from Dec 1, 2026)$4.00$0.20$20.00

These figures come from GitHub's Models and pricing documentation. They are per million tokens.

Today, Neon's per-token price is higher than Carbon's promotional price. It drops below Carbon's list price once that promotion ends on December 1. The claim of a lower cost per task therefore depends on how many tokens each model uses to finish a job. Google's pitch is that Neon needs fewer tokens or less rework. Per-token pricing alone doesn't show that, and I can't verify it.

Billing is token-based. GitHub's docs say each interaction consumes input, output and cached tokens that are converted to AI credits, where one credit equals $0.01. Long agent sessions that read lots of context can therefore add up quickly.

What administrators need to know​

For Business and Enterprise, the key detail is the default. GitHub says under default model enablement, new models are automatically enabled unless an administrator has turned off the global default or explicitly disables this model. If you've done nothing, Neon will appear for your developers once the rollout reaches them.

GitHub's enterprise documentation describes the controls:

  • Enterprise owners can enable or disable a model for everyone, or delegate the decision to organizations.
  • Models you haven't configured show as "Delegate to Default Policy". The "Default availability" policy decides whether they are on or off.
  • If you don't like that behavior, you can disable specific models, limit them to certain organizations or teams, or switch off default availability so unconfigured models start disabled.
  • At the organization level, GitHub notes that a setting enforced by the enterprise cannot be overridden by an organization owner.

An independent commentary on the earlier Carbon release made the same point about the opt-out default. It warned that model rollouts can move faster than organizations review them, which is where usage-based charges can build up. That is an opinion from a third party, but it's a fair one.

A sensible approach is to decide deliberately whether Neon belongs in your approved list. Check your budget controls before you encourage heavy use in agent sessions. GitHub's docs describe a budget option that can stop usage when a limit is reached.

Context: a crowded Gemini lineup​

Neon is the fourth recent Gemini 4 model in Copilot's documentation. Carbon arrived on November 10 and is designed for complex coding, agentic, and reasoning tasks, with a focus on secure code generation. Hydrogen followed on November 13 as a fast Flash model. Nothing in GitHub's announcement says Neon replaces Carbon, so think of it as one more option rather than a successor.

One loose end: before the announcement, a tech site reported only an unconfirmed code reference to a Neon model string in Google's Gemini web app. It added that Google's public models page listed no Neon endpoint at the time. That report predates GitHub's launch and doesn't change what GitHub has now announced. I didn't verify Google's own developer documentation for Neon.

Bottom line​

  • Neon is generally available in Copilot for Pro+, Max, Business and Enterprise, with a gradual rollout.
  • The speed, cost-per-task and accuracy advantages over Carbon are Google's claims. Neither GitHub's changelog nor the pricing documentation backs them with numbers.
  • Per-token pricing is $3 input and $15 output per million tokens, so real savings depend on how many tokens a task uses.
  • Business and Enterprise admins should review model policy and budgets now, because new models are on by default.

The practical test is simple. Run Neon and your usual model on the same few tasks, then compare the AI credits used and the quality of the results.

 

References

  1. Gemini 4 Neon in GitHub Copilot GitHub Changelog 2026-11-18T21:49:21+00:00
  2. Models and pricing for GitHub Copilot - GitHub Docs docs.github.com
  3. Configuring access to AI models in GitHub Copilot - GitHub Docs docs.github.com