Google has released Gemini 4 Neon, a new member of its Gemini 4 family of AI models aimed at coding, analysis and business tasks that run over many steps. The model went on general release on 18 November (US time) and costs $3 per million input tokens and $15 per million output tokens, with room for up to 1 million tokens of context.
Google places Neon in the middle of the family. In its description, Gemini 4 Carbon is the model for the hardest, longest projects, Hydrogen is the fast option for everyday jobs, and Neon is meant to strike a balance between the two.
"Neon is built for complex coding, analytical reasoning, and production-grade enterprise workflows."Google, Gemini 4 Neon announcement
The details
- Thinking levels: Neon offers four levels of reasoning effort. "High" is the default and "xhigh" is the most thorough.
- Agents: inside Google Antigravity, the company's coding environment, Neon can coordinate several Hydrogen models working in parallel as helpers.
- Inputs: according to Google Cloud's documentation, the model takes text, images, documents such as PDF and Office files, up to about 8.4 hours of audio and up to about 45 minutes of video, and replies in text. Its knowledge runs to March 2026.
- Cost per task: Google says that at the default setting Neon completes work for 32% less than Carbon, and for roughly 50% to 80% less than leading rival frontier models on two of the indexes it cites.
Who gets it and when
Google says Neon is rolling out now in these places:
- Developers: the paid tier of the Gemini API in Google AI Studio, and Android Studio.
- Businesses: through the API and the model menu in Gemini for business.
- Consumers: Google AI Pro and Ultra subscribers can choose it from the model menu in the Gemini app, and paid Pro and Ultra subscribers can use it in Antigravity.
- Security teams: a version called "Gemini 4 Neon for cyber" is due to reach vetted defenders in the coming days through Google's Fairwind Program.
On Google Cloud the model is listed as generally available under the ID gemini-4-neon, served from global, US and EU locations. The announcement does not name any countries for the app rollout or mention India.
Numbers
Google published a long list of test scores. A selection:
| Test | Gemini 4 Neon |
|---|---|
| Artificial Analysis Intelligence Index (xhigh) | 61, third overall |
| Artificial Analysis Coding Agent Index | 58.6, first overall |
| Vals Index | 70.9%, second |
| Terminal-Bench 4.1 | 65.2% |
| Zapier AutomationBench | 56.3%, first overall |
| SEC-bench Pro (security) | 90.5% |
On safety, Google says Neon keeps the same protections against serious misuse and prompt injection as Carbon while refusing far fewer legitimate requests. It reports a misaligned action rate of 0.09% on its agent safety tests and a 99.2% pass rate on an honesty test for agents. In a coding test designed to catch models taking shortcuts, Neon did so on 2.45% of tasks, which Google says is the lowest of any Gemini 4 model.
What is still unknown
The scores come from Google and the test makers it cites; independent results will take time. The announcement gives no list of countries for the Gemini app rollout and no date for the cyber version beyond "the coming days". Google points to a separate model card for fuller safety details. Our earlier Gemini coverage includes file editing in the Gemini app.
Announced by Google on its official blog: read the announcement







