Google began rolling out Gemini 4 Neon on November 18, 2026, pitching it as the middle option between Hydrogen and Carbon for coding, analysis and knowledge work. The periodic table has apparently joined the product team. Source
Google AI Pro and Ultra subscribers can select Neon in the Gemini app as the rollout reaches them, with access also coming to paid subscribers in Google Antigravity. Developers get access through the Gemini API paid tier in AI Studio and Android Studio, while enterprise access includes Gemini for business through its API or model dropdown. API usage rates are separate from consumer subscription eligibility. Source
According to 9to5Google, Neon replaces Argon and 3.1 Pro in the app and takes the “Deeper analysis” slot. Hydrogen remains “Everyday help,” while Carbon is labeled “Code & complex work.” At least the menu translates the chemistry into actual jobs. Source
The cheaper model depends on the date
Google announced Neon pricing of $3 per million input tokens and $15 per million output tokens, with cached input priced at $0.15 per million tokens. Source
Carbon’s promotional token rates remain lower through November 30, according to 9to5Google. All prices below are per million tokens. Source
That makes Neon 50% more expensive per input or output token than Carbon’s current promotional rates, but 25% cheaper than Carbon’s reported December rates. Source Source
What Neon brings to the middle
Neon has a 1 million-token context window, the amount of material it can work with at once. Google says it can synthesize information across spreadsheets, PDFs, charts, audio and video. There are also four thinking levels, which 9to5Google identifies as Low, Medium, High and Extra High, to adjust reasoning effort and token spending. Source Source
Google says Neon can handle long-running coding workflows on its own or direct parallel Hydrogen agents in Antigravity. In the Gemini app, the company pitches it for work such as turning dense uploads into reports, researching current information and checking billing statements or receipts. Less glamorous than building an AI demo, perhaps, but finding the mysterious charge is a perfectly respectable assignment. Source
Google reports a score of 61 and third place on the Artificial Analysis Intelligence Index at Neon’s highest, xhigh thinking setting. Separately, it claims a 32% lower cost per task than Carbon at the default high setting. That task-cost result doesn’t establish savings on every workload, and it isn’t a reduction in the posted token rates. Source
Comments