Google Releases Gemini 4 Neon: 3x Faster Than Carbon, Ranks Third on AA Index with Score of 61

·2026.11.19 08:08

Key point

Available immediately across Google AI Studio and the Gemini app, the model offers 60% of Carbon's cost while maintaining comparable quality in early evaluations.

Details

  • Google launched Gemini 4 Neon, an adaptation of Carbon, positioning it as faster and cheaper than Carbon but smarter than Hydrogen, with immediate availability across paid API tiers, enterprise services, and the Gemini app for Pro and Ultra subscribers.
  • Neon ranks third overall on the Artificial Analysis Intelligence Index v5.01 with a score of 61, trailing only two higher-scoring models, while ranking first on the AA Coding Agent Index with a score of 58.6.
  • Early evaluation by Glean found Neon achieves quality equal to or better than Carbon at three times the speed and 60% of the cost, with API pricing set at $3 per million input tokens and $15 per million output tokens.
  • The model supports a 1 million token context window and offers four reasoning depth levels (Low, Medium, High, Xhigh), with a knowledge cutoff of March 2026.

Performance and Pricing

  • Neon scores 70.9% on the Vals Index (2nd place) and leads in Legal Research Bench, Tax Agent Bench, and AutomationBench (56.3%).
  • Input and output prices are 62.5% lower than GPT-6.2 Astra and 25% lower than Claude Opus 5.5, with cached input priced at $0.15 per million tokens.
  • Google claims Neon costs 32% less per task than Carbon at the default high setting and sits on the Pareto frontier for efficiency.

Safety and Cybersecurity

  • Neon achieved a 0.0% attack success rate on Gray Swan's indirect prompt injection benchmark when using the highest thinking setting with real-time safeguards.
  • The model recorded a 2.45% shortcut rate in coding-task cheating evaluations, the lowest among Gemini 4 models, and a 99.2% pass rate on agent-honesty evaluations.
  • Cybersecurity benchmarks include 76% on the hard tier of AutoBaxBench and 90.5% on SEC-bench Pro for vulnerability discovery.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.