AI news, decoded · 5 minutes

Gemini 4 Neon Is Out: Index 61, #1 for Coding Agents

On November 18, 2026, Google released Gemini 4 Neon, the fourth Gemini 4 model in seven weeks, slotting between Carbon and Hydrogen: complex coding, analytical reasoning and scientific work — the cost-effective pick for frequent, hard tasks. It is live today on the Gemini API paid tier (AI Studio, Android Studio), Gemini for business, Antigravity (Pro / Ultra subscriptions) and the Gemini app (Pro / Ultra subscribers). The announcement says nothing about a free tier, so do not count on one. Context is 1 million tokens.

Where it is strong

  • Scores: on the independent Artificial Analysis Intelligence Index (xhigh, the highest reasoning level), Neon scores 61, third overall — behind only Claude Fable 5.2 at 65 and GPT-6.2 Astra at 62, and 1 point ahead of Google's own Carbon at 60.
  • #1 for coding agents: it tops the Artificial Analysis Coding Agent Index at 58.6; Terminal-Bench 4.1 reaches 65.2% and FrontierCode 2.0 reaches 46.2% (at the default high reasoning level).
  • Strong at knowledge work: #2 on the Vals index (70.9%), #1 on both the Legal Research and Tax Agent benchmarks; an Agent Arena debut at #2 (1336 points), with steerability also #2.
  • Price: $3 per million input tokens and $15 per million output, cached input $0.15 (95% off). That is 25% cheaper than Carbon's standard $4/$20 — but Carbon's $2/$10 promotional price runs until November 30, so during the promo Neon is actually 50% more expensive than Carbon; do not get that math backwards. On Artificial Analysis's measure, Neon costs about $6.44 per task, roughly 17% less than Carbon's $7.76; on Google's measure, the default high level costs 32% less per task than Carbon.

Hold the applause

  • The 61 is at xhigh, the highest reasoning level; the default level scores a notch lower. Google's own Vals and Agent Arena numbers are likewise unverified, and so far Artificial Analysis is the only third party with a full score.
  • No free entry point: the Gemini app and Antigravity are limited to Pro / Ultra subscribers and the API is a paid tier; Google has not said when free users or AI Mode in Search will get Neon.
  • It is a middle tier, not a new flagship: the hardest long-horizon projects are still Carbon's job. Neon's pitch is frontier-level ability on frequent, hard tasks — fast and cheaper — not beating the flagship across the board.

Also that day: the Genie 4 world model

The same day, DeepMind put Genie 4 into the new Project Genie: for the first time you can edit the world with text as you walk (change the weather, add characters, alter the plot), with native real-time audio (ambient sound, music and character dialogue that follow the scene), and generated content carries a SynthID watermark. But the bar is high: it is open only to eligible AI Ultra subscribers, with Pro users waitlisted; Google itself admits long-session consistency, complex physics and audio accuracy are still being worked on — Genie 4 is a frontier research prototype, not a finished product.

Sources