Google’s new model for hard software, knowledge and science work: what it does, where to use it, and what it costs
7 bite-sized sections
2 min read
Free, no signup
Short answer
A free Google Gemini 4 Neon guide as of November 21, 2026: what Google released on November 18, where Neon fits between Gemini 4 Carbon and Gemini 4 Hydrogen, its 1 million token context and output limits, the pricing Google published at launch, where you can use it (Gemini API, AI Studio, Antigravity, Gemini for business and the Gemini app for paid plans), the benchmarks Google reported, its safety claims, and when to pick Neon instead of Hydrogen.
Gemini 4 Neon is a Google model released on November 18, 2026. Google positions it for software engineering, knowledge work, science, complex coding, analytical reasoning and enterprise workflows. It sits between Gemini 4 Carbon, for the hardest long projects, and Gemini 4 Hydrogen, for fast everyday work.
Hydrogen for the quick jobs, Neon for the hard ones, Carbon for the longest projects.
02
Context and output
Input context
1 million tokens
Output
Up to 64K tokens by default for normal requests; Google’s model card also mentions up to 1 million through background long decoding
Thinking levels
Google names high (the default) and xhigh in its launch post; check the model card for the full list
Knowledge cutoff
March 2026, per the model card
03
Pricing at launch
Input
$3 per million tokens
Cached input
$0.15 per million tokens
Output
$15 per million tokens
This is Google’s published launch price. Check Google’s pricing page before you budget, since prices can change.
04
Where you can use it
Developers
Gemini API, with the paid tier in Google AI Studio and Android Studio
Businesses
Gemini for business, through the API or the model dropdown
Antigravity
For paid Google AI Pro and Ultra subscribers
Gemini app and AI Studio
For Pro and Ultra subscribers, through the model dropdown
Security defenders
A Fairwind Program for cybersecurity defenders, by request
Google’s launch material does not mention Search or Workspace availability for Neon. Check before you plan around them.
05
What Google reported
Artificial Analysis Intelligence Index: 61 at xhigh, ranked third by Google’s citation
Terminal-Bench 4.1: 65.2%
HLE-Diamond: 71.3%
Google says Neon trails Claude Opus 5.5 on Terminal-Bench 4.1 and FrontierCode 2.0 in its model card
These are Google’s own figures from its launch post and model card. Test on your own work.
06
Safety claims
Google says Neon keeps the protections of Carbon against severe misuse and prompt injection
Reported misaligned action rate of 0.09% and agent honesty pass rate of 99.2%
Fewer unjustified refusals than Carbon, per Google’s internal evaluations
Google’s full safety report was marked as coming soon when we checked
07
When to pick Neon
Large codebase changes that need careful reasoning
Analysis of long, complex documents or data
Agent workflows where mistakes cost more than tokens
Quick edits and routine tasks: use Hydrogen instead
The golden rule
Do not use the most powerful model for every task. Use Neon when the work is hard and the cost of a wrong answer is high, and check its output before you ship it.
Neon
Carbon
Hydrogen
1M context
Pricing
Benchmarks
Safety
Run one hard task you already know the answer to through Neon and compare it with Hydrogen.
For research and ideas only. Results vary and nothing here is a promise of income. See our Disclaimer.
See you Monday!
Weekly newsletter
Everything new, every Monday.
One short email every Monday with everything shipped that week: new free tools and websites, new guides, new free browser tools and what's new in my free Gumroad libraries, plus how the site is growing. Free, and you can unsubscribe anytime.
Visits are counted by Cloudflare Web Analytics, which sets no cookies and stores nothing on your device. An affiliate tracking script from Impact.com also loads on our pages and can set its own cookies so partner companies can credit us for sign-ups. Essential storage is always on so pages load and work correctly. You decide whether your browsing preferences are saved on your device. Read the Privacy Policy.