Google has released Gemini 4 Neon, a new model in its Gemini 4 family. It sits between the high-end Gemini 4 Carbon and the faster, cheaper Gemini 4 Hydrogen. For Android users, the main point is access: Neon is rolling out now to both Google AI Pro and AI Ultra subscribers in the Gemini app. That means people on the cheaper Pro plan get a frontier-class model without paying for Ultra.
Google DeepMind SVP and Google Chief AI Architect Koray Kavukcuoglu announced the model on November 18. Google pitches it as offering “frontier-level intelligence with fast end-to-end task completion” across software engineering, knowledge work and science.
Where Neon fits in the Gemini 4 lineup

Neon is the fourth Gemini 4 model in a few weeks. It follows Gemini 4 Argon and Gemini 4 Carbon, which Google made broadly available on November 10, and Gemini 4 Hydrogen, which launched November 12. Google calls Carbon its model for the hardest problems. Hydrogen is the fast, low-cost model for everyday agentic work.
Neon aims for the middle ground. Developers pay $3 per million input tokens and $15 per million output tokens. Cached input costs $0.15 per million tokens, which Google says is a 95% discount. Like its siblings, Neon has a 1 million-token context window.
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Gemini app access at launch |
|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $7.50 | Rolled out across the Gemini app |
| Gemini 4 Neon | $3 | $15 | AI Pro and Ultra subscribers |
| Gemini 4 Carbon (introductory, through Nov. 30) | $2 | $10 | AI Ultra subscribers |
| Gemini 4 Carbon (from Dec. 1) | $4 | $20 | AI Ultra subscribers |
Through the end of November, Carbon's introductory per-token price is actually lower than Neon's. Even so, Google says Neon at its default “high” thinking level costs 32% less per task than Carbon. When Carbon's introductory period ends on November 30, its price rises to $4 input and $20 output per million tokens, and Neon becomes the cheaper option per token as well.
Benchmarks and coding claims
Google says Neon scored 61 on the Artificial Analysis Intelligence Index at its highest “xhigh” thinking level. That puts it third overall, behind larger and slower models. Google's other headline numbers include:
- #2 on the Vals Index (70.9%), and #1 on the Legal Research Bench and Tax Agent Bench
- #1 on the Artificial Analysis Coding Agent Index (58.6), with 65.2% on Terminal-Bench 4.1
- A score of 1336 on Arena's new Agent Arena Index, which ranks it #2 globally and also #2 on steerability, meaning how well it handles corrections mid-task
- #1 on Zapier's AutomationBench (56.3%) and on Andon Labs' Vending-Bench 2 business simulation ($16,843)
- Best in class on Terminal-Bench Science (70.5%), RiemannBench (84.0%) and HLE-Diamond (71.3%)
All of these figures come from Google's announcement. Google also says Neon recorded the company's lowest fabrication rate yet in internal testing in Google Antigravity, its agentic coding tool.
For developers, Google says Neon is built for long-running agent work. It can work alone or act as an “Orchestrator” that directs parallel Hydrogen subagents in Antigravity. On the SWE-sweep benchmark, Google says Neon scored 12.8% on the mini-SWE harness and 39.0% inside Antigravity.
Neon has four thinking levels so teams can trade intelligence against token spend. Google says that at the default high setting, Neon costs roughly 50% to 80% less per task than top competing frontier models. At xhigh, Google says it still costs at least 36% less per task than rivals at their highest settings.
Security tools and safeguards
Google is also preparing a cybersecurity version, Gemini 4 Neon for cyber. It will reach trusted defenders in the coming days through the company's Fairwind Program. Google says it writes more secure code than Carbon on AutoBaxBench and scored 76% on that benchmark's hard tier. It also reached 90.5% on SEC-bench Pro for finding vulnerabilities and 70.7% on Collinear's CWE-bench v2 for patching them.
On safety, Google says Neon keeps the same protections against severe misuse and prompt injection as Carbon. Google also says Neon turns down far fewer legitimate requests in sensitive areas, refusing 1.98% of tasks on Terminal-Bench 4.0 and none on Terminal-Bench-Science.
Availability
As of November 19, 2026, Neon is rolling out in the Gemini API in Google AI Studio and in Gemini for business, through both the API and the model dropdown. Paid Google AI Pro and Ultra subscribers can also use it in Google Antigravity and in the Gemini app. Pro and Ultra subscribers on Android should see it in the Gemini app's model selector as the rollout reaches their accounts.
Google hasn't said whether Neon will come to free Gemini users or to AI Mode in Search, where Carbon and Hydrogen are already available. The next date to watch is November 30, when Carbon's introductory API pricing ends. Neon for cyber should reach Fairwind Program participants in the coming days.