Gemini 4 Neon is Google's new middle model in the Gemini 4 family, adapted from Gemini 4 Carbon and priced below it for complex, frequent agent work. For most teams it is a routing decision, not a straight swap: move tool-heavy, repeated tasks from Carbon to Neon, keep Hydrogen for volume, and test before you switch.
What was released
Gemini 4 Neon was announced by Google on 18 November 2026 and began rolling out the same day. Google places it between Carbon, which handles the hardest long-horizon projects, and Hydrogen, which handles fast everyday execution. In Google's words, it brings "frontier-level intelligence with fast end-to-end task completion".
Access by channel, according to the announcement:
- Gemini API: the paid tier, through Google AI Studio and Android Studio. The AI Studio link in the announcement uses the identifier
gemini-4-neon. - Gemini for business: through the API or the model dropdown.
- Google Antigravity and the Gemini app: for Google AI Pro and Ultra subscribers.
- GitHub Copilot: generally available the same day on Copilot Pro+, Max, Business and Enterprise, across VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI and the coding agent. It is billed at provider list pricing under usage-based billing, and Business and Enterprise admins control it through the model policy.
- A cyber variant: rolling out to vetted security teams in Google's Fairwind Program.
The announcement does not name Vertex AI, and it does not say whether the API model is a preview or a stable release.
What changed compared with Carbon and Hydrogen
The Gemini 4 Neon model card and the announcement give the facts that matter for an engineering team:
- Same base as Carbon: Neon is adapted from Gemini 4 Carbon, and the model card points to Carbon's card for architecture, training data and acceptable use.
- Lower price than Carbon: Google lists separate rates for input, cached input and output in the announcement, with cached input at a small fraction of the full input rate, and reports a lower cost per task than Carbon at the default thinking level.
- Same context and output limits as Hydrogen: text, image, audio and video input within a shared 1M-token input and output window, up to 64K output tokens by default for a synchronous call, and longer outputs through background long decode.
- Four thinking levels: high is the default and xhigh is the top setting.
- Fewer unjustified refusals than Hydrogen: the model card reports fewer over-refusals on benign prompts than Hydrogen, particularly on multimodal prompts, with other content safety results within confidence intervals.
- Knowledge cutoff of March 2026, and the full safety and security report is still to come.
Google also says Neon reads redundant files less often and compacts context during long runs, and in Antigravity it can act as an orchestrator that hands work to parallel Hydrogen subagents. That last point is an Antigravity feature, not an API capability. The benchmark figures in the announcement are Google's own; treat them as a reason to test, not as a result.
What it means for teams building AI agents
Treat it as a third routing tier. Many teams run two models: an expensive one for planning and hard reasoning, and a cheap one for volume. Neon fills the gap between them. Candidates to move from Carbon are tasks that need real reasoning but run many times a day: code review on every pull request, multi-step document analysis, and tool-heavy support workflows. Keep Carbon for the long, open-ended tasks where a wrong plan is expensive, as we set out in our notes on Gemini 4 Carbon and Argon.
Run the same eval set on all three. Neon is a new model, so behaviour on your prompts will move. Run the test set you use today against Carbon, Neon and Hydrogen and compare pass rate, tokens per task and latency at each thinking level. Route each task type to the cheapest model and level that passes. Our guide to what AI agent evals catch covers how to build that set.
Lean on caching for long agent runs. The cached input rate is a small fraction of the full input rate, so agents that resend the same system prompt, tool schemas and documents on every step gain the most. Check that your prompt layout keeps the stable parts first so they can be cached, and track cost per completed task rather than cost per call, as described in controlling AI agent running costs.
Revisit refusal handling on image and document flows. If you moved image-heavy work off Hydrogen because of refusals, Neon is worth a retest, since the model card reports fewer unjustified refusals there. Measure refusal rate on its own, not only overall accuracy.
Check output length and governance before rollout. Flows that write long reports or large files in one call can still hit the 64K synchronous default. On the governance side, if your developers use GitHub Copilot, Neon may already be enabled under default model enablement, so decide whether it belongs in your approved model list.
When not to switch yet. Wait if your compliance team needs the full safety and security report before approving a model, if you run on Vertex AI, which the announcement does not name, or if your traffic sits on the Gemini API free tier, which Neon is not open to. Keep the model ID in configuration either way: our model deprecation notes explain why that makes the next change cheap.
Frequently asked questions
What is the model ID for Gemini 4 Neon?
The Google AI Studio link in Google's announcement uses gemini-4-neon, which points to that identifier for the Gemini API. Confirm the exact ID in the AI Studio model list for your project before hard-coding it. Keep it in configuration rather than code, so you can route tasks between Neon, Carbon and Hydrogen, or roll back, without a deployment.
Is Gemini 4 Neon available on the Gemini API free tier?
No. Google says Neon is available on the paid tier of the Gemini API in Google AI Studio and Android Studio, and to Google AI Pro and Ultra subscribers in AI Studio, the Gemini app and Antigravity. That differs from Hydrogen, which Google opened to all API tiers. Projects testing on the free tier need billing enabled before they can call Neon.
How is Gemini 4 Neon different from Gemini 4 Carbon?
Neon is adapted from Carbon and shares its architecture and training data, according to the model card. Google positions Carbon for the hardest long-horizon projects and Neon for complex work that runs often, at a lower price per token and a lower cost per task. Test both on your own tasks, because the right choice depends on how costly a weak plan is.
Can I use Gemini 4 Neon in GitHub Copilot?
Yes. GitHub made Neon generally available on 18 November 2026 for Copilot Pro+, Max, Business and Enterprise, in VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI and the coding agent. It is billed at provider list pricing under usage-based billing. Business and Enterprise admins can turn it on or off through the Copilot model policy.



