Gemini 4 Neon is live. Google has launched it across the Gemini API, Gemini for Business, Google Antigravity, and the Gemini app — replacing Argon as the model powering “Deeper analysis” in the consumer app, according to 9to5Google’s reporting. It’s available now to Google AI Pro and Ultra subscribers in the US and India, with no regional pricing differences cited in the sources.
In India, AI Pro costs ₹2,499/month. In the US, it’s $19.99/month, same as the newly named Gemini 4 Neon subscription tier announced last month. But Neon itself isn’t a standalone plan. It’s a model, one that slots between Gemini 4 Hydrogen (everyday help) and Gemini 4 Carbon (complex work), sitting third in Google’s current four-model hierarchy.
It follows Carbon, which launched 10 November, and Hydrogen, which arrived two days later. Argon, the first Gemini 4 model, debuted 30 September. Neon’s price: $3 per million input tokens, $15 per million output tokens, with cached input at $0.15. Its context window is 1 million tokens.
It offers four reasoning levels, low, medium, high (default), and xhigh, and scored 61 on the Artificial Analysis Intelligence Index at xhigh, trailing GPT-6.2 Astra and Claude Fable 5.2 by one point each, and edging ahead of Carbon by one. That edge comes at a cost trade-off. Neon uses about 15% more output tokens per task than Carbon, but its lower token pricing brings the average task cost down to $6.44 versus Carbon’s $7.76, unless you’re using Carbon’s introductory pricing ($2/$10), which runs through 30 November.
Under that deal, Neon is 50% more expensive per task, Artificial Analysis noted. In coding agent benchmarks, Neon scored 59, matching Claude Opus 5.5 and beating Fable 5.2, while costing $6.86 per task, less than half of Opus ($12.61) and Fable ($15.47). A cyber-focused version is rolling out “over the next few days” to eligible defenders via Google’s Fairwind Program, the same channel that delivered Argon to security teams first.
That variant scored 90.5% on SEC-bench Pro, Google says, at roughly half the cost per challenge of rival models. It retains Carbon’s safeguards against cyber and biosecurity misuse, but Google claims it rejects far fewer legitimate requests in those domains. Neon also acts as an orchestrator in Antigravity, directing parallel Hydrogen subagents, a new capability Google highlighted.
No launch date was given for broader enterprise or developer access beyond what’s already live. But rollout is underway, and the model is already active for paying subscribers who select it in the Gemini app or configure it in Antigravity.
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.