Google released Gemini 4 Neon on Nov 18, positioning it as a model built for long-running agent sessions, with four thinking levels and a 1M token context window, priced at $3 per million input tokens. Early hands-on reviews are split: Fast.ai founder Jeremy Howard and the YouTube channel Prompt Engineering gave positive assessments, while developer Simon Willison criticized Neon for making the already complicated Gemini 4 pricing structure even more confusing.

Hands-on performance: long sessions and complex tasks win praise

In its video, Prompt Engineering said it received early access and demonstrated projects Neon actually built, including an on-call agent running on Google Cloud Run. The video description includes an AI Studio trial link and a Google developer guide, and the author disclosed that he recently joined Google and that the views are his own.

Fast.ai founder Jeremy Howard posted on X that after trying the Neon feature in the Gemini App, he confirmed it performs "as good as everyone says." The endorsement from a well-known figure in deep learning further supports the positive word of mouth around Neon.

Simon Willison, meanwhile, used his usual "pelican test" to compare Neon's low, medium and high reasoning levels, adding the results to his Gemini 4 comparison grid. He described Neon's pelican output as "looks better than Hydrogen, worse than Carbon." He also used the "extra" effort level in Google Antigravity to have Neon generate a Three.js animation in which "Yonkers stands up and turns into a giant mecha, stomping through Manhattan, Queens and Brooklyn," requiring the use of OpenStreetMap map data.

Pricing controversy: official chart "pretends Argon doesn't exist"

Simon Willison noted that the Gemini 4 lineup previously had three models — Hydrogen, Argon and Carbon — with Argon and Carbon at the same price and Hydrogen currently at three-quarters of that price, set to become one-quarter once the initial discount expires on Dec 1 and the other two double in price. Yet Neon's launch materials "pretend Argon doesn't exist," with the official new chart listing only Carbon, Neon and Hydrogen.

Neon has somehow made this even worse.

According to Willison's breakdown, list prices from highest to lowest should be Carbon, Neon, Hydrogen. But because Carbon is currently at half price, the effective current prices put Neon highest ($3 input/$15 output), followed by Carbon (2/10) and Hydrogen (1.50/7.50). He expects the pricing logic to make more sense only after Carbon doubles to 4/20 on Dec 1.

ModelList price rankingCurrent effective price (input/output, USD per million tokens)Price after Dec 1 (input/output, USD per million tokens)
CarbonHighest2/10 (half price)4/20
NeonMiddle3/153/15
HydrogenLowest1.50/7.501.50/7.50

What to watch: Dec 1 price adjustment is the key moment

For now, praise for Neon's capability and the pricing controversy are running in parallel. Supporters focus on its long-agent-session positioning and 1M context, while critics focus on the confusion over product-line naming and price structure. After Carbon's price doubles on Dec 1, the effective price ranking of the Gemini 4 series will change, and it is worth watching whether Neon remains the most expensive tier and whether Google clarifies Argon's positioning.