Skip to content
Neural Daily

ModelsNews

Gemini 4 Neon: 61 on the AA Index, first in coding

🇭🇺 Olvasd el magyarul →

Google released Gemini 4 Neon, a new standalone frontier model in the Gemini 4 family, not a Hydrogen update. On the Artificial Analysis Intelligence Index it scores 61 at xhigh thinking and ranks 3rd; on the Coding Agent Index it is first with 58.6. It costs $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window, and starts rolling out today.

· 4 min read

A Gemini 4 Neon hivatalos bejelentő grafikája: narancssárga neonfelirat sötét falon
Image: Google · Press kit · Official blog header from deepmind.google · Source

What's new?

Google introduced Gemini 4 Neon on November 18 (Google DeepMind). It is a new standalone model, not a Hydrogen update. In the family, Carbon is for the hardest, long-horizon projects, Hydrogen is for fast everyday execution, and Neon balances depth and efficiency.

  • Artificial Analysis Intelligence Index: 61 at xhigh thinking, 3rd overall, just behind larger, slower models.
  • Artificial Analysis Coding Agent Index: 58.6, 1st overall.
  • Vals Index: 70.9%, 2nd, on the Pareto frontier.
  • Price: $3 per million input tokens ($0.15 per million cached input tokens, a 95% discount) and $15 per million output tokens, with a 1 million-token context window.

The numbers

Google says Neon is strong in software engineering, knowledge work, and science workflows. It scores 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at its default high setting. It also ranks first on the Vals Legal Research Bench and Tax Agent Bench.

Further figures from Google's announcement:

  • Terminal-Bench Science: 70.5%. RiemannBench: 84.0%. HLE-Diamond: 71.3%. Best in class on all three.
  • Zapier AutomationBench: 56.3%, first overall. Andon Labs Vending-Bench 2: $16,843, also first.
  • Agent Arena Index: 1336, 2nd globally, across 4,200+ real-world agent sessions. On the same index it also ranks 2nd for steerability: it adapts cleanly to mid-task user corrections.
  • Chartography: 73.5%. Agents' Last Exam v1.1: 75.8%, on par with leading models.

At its default high thinking level, Google says Neon costs 32% less per task than Carbon. It sits on the Pareto frontier on both the Vals Index and Agent Arena at high, with roughly 50-80% lower cost per task than top competing frontier models. xhigh is its highest intelligence score, and still costs at least 36% less per task than competing models at their highest setting.

Neon has four thinking levels, so teams can match intelligence and token spend to the workload.

Orchestrator: Neon directs, Hydrogen executes

Neon can run standalone, or act as a fast orchestrator in Antigravity directing parallel Hydrogen subagents (Gemini 4 Hydrogen: Argon-level smarts at $1.50 input). Google says it cuts down on redundant file reads, shares progress as it works, and handles context compaction reliably in long sessions. We previously covered Antigravity /teamwork, where Gemini 4 agents take on open problems (Antigravity /teamwork: Gemini 4 agents take on open problems).

Safety

Google says Neon keeps Carbon-level protections against severe misuse and prompt injection. On the Gray Swan IPI benchmark, at its highest thinking level with real-time safeguards, the attack success rate was 0.0%. In internal Antigravity testing it recorded the lowest fabrication rate yet among Gemini models.

Where to get it

Neon starts rolling out today in the Gemini API in Google AI Studio, in Gemini for business (via the API or the model dropdown), in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers. Developers can also use the paid Gemini API tier in Android Studio, and every Pro and Ultra subscriber can try it in AI Studio. The Neon for cyber variant will reach eligible Fairwind Program members over the next few days.

Why it matters

If you run coding or analysis agents and Carbon is too expensive or slow for day-to-day work, Google positions Neon as frontier-level capability at a lower cost per task: 32% cheaper than Carbon at high. The $3 input price sits between Hydrogen's $1.50 and the larger models, and cached input is $0.15. A practical first step is to try it on the coding work you currently give Hydrogen or Carbon, and to see whether Neon works better as an orchestrator with Hydrogen subagents.

    By Mikola Máté Patrik

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles