Gemini 4 Neon is now generally available — not as a standalone app or consumer-facing feature, but as a new mid-tier AI model in Google’s Gemini 4 family, priced at $3 per million input tokens and $15 per million output tokens. It lands between Gemini 4 Carbon ($4/$20) and Gemini 4 Hydrogen ($1.50/$7.50), targeting users who need more reasoning depth than Hydrogen offers but don’t require Carbon’s full long-horizon research capacity.
The launch was confirmed by Google and reported by GSMArena on 19 November 2026. It is not a phone, tablet, or wearable. It belongs in AI models, not mobiles. Section corrected.
Neon is accessible today in three places: the Gemini API via Google AI Studio, Gemini for Business, and Google Antigravity, but only for paid Google AI Pro and Ultra subscribers. Those users can now select it from the model dropdown inside the Gemini app. It has a 1 million-token context window.
It supports four thinking levels, xhigh, high, medium, low, with xhigh reserved for the most complex tasks. At its default high setting, Google says Neon costs 32% less per task than Carbon. In benchmark comparisons shared by Google, Neon outperforms Claude Opus 5.5 and GPT-6.2 Sol on the Artificial Analysis Intelligence Index (61 vs. 59 and 57) and the Vals Index (70.9% vs. 70.1% and 66.1%).
Opus 5.5 leads narrowly on Terminal-Bench 4.1 and FrontierCode 2.0, but those are the only two tests where it does. Google also claims Neon delivers its lowest rate of factual inaccuracies, “lies”, in internal testing so far. A security-specific variant, Gemini 4 Neon for cyber, is slated for rollout “in the coming days”, but only to vetted defenders enrolled in Google’s Fairwind Program.
No public release date, no pricing, no broader availability announced. This isn’t a device. It’s infrastructure. And it’s already live, for developers, enterprises, and paying subscribers.
The next step? Wider integration into Google Cloud tools, deeper Antigravity orchestration, and, if early benchmarks hold, pressure on Anthropic and OpenAI to sharpen their mid-tier positioning. No Indian or US device pricing applies. No launch event.
No hardware. Just code, context, and cost-per-token, now active.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.