Google equips Gemini with a faster model for complex tasks
Google rolls out Gemini 4 Neon for AI Pro and AI Ultra subscribers. The model promises a level close to Carbon, but with faster responses and lower cost.
Google has launched Gemini 4 Neon, an AI model presented as capable of reaching a reasoning level close to Carbon while remaining faster and less costly to use. The launch initially targets AI Pro and AI Ultra subscribers within the Gemini app, replacing 3.1 Pro and Argon. In practice, Google aims to better allocate its models between daily tasks, advanced analysis, and very complex work.
A new balance between speed and depth
Neon officially sits between Hydrogen and Carbon in the Gemini 4 family. Google describes it as suitable for complex and frequent workflows, with performance comparable to, or even exceeding, Carbon on code, reasoning, and agent tasks, but at Flash speed and for a lower cost.
The positioning is quite clear: Carbon remains reserved for the heaviest projects, Hydrogen for quick daily uses, and Neon takes the middle ground. It’s not very glamorous, but this is often where true enterprise adoption happens.
What Google promises in the Gemini app
In the Gemini app, Google highlights several concrete uses. Neon is expected to produce better-structured texts, read imported files more precisely such as spreadsheets, PDFs, and photos, then leverage relevant data from Google tools like Gmail and Docs.
The company also emphasizes search and analysis. According to the launch note, Neon “incorporates details from sources across the web” to keep responses up to date on shifting topics, such as price comparisons between suppliers or preparing a brief with recent data.
Another promise: speeding up multi-step tasks. Google notably cites the analysis of billing statements, receipt comparisons, or preparing a reimbursement request. Here, the benefit isn’t spectacular but very concrete: saving time on very real chores.
A model designed for professional use
Neon maintains a context window of one million tokens and four levels of reflection: Low, Medium, High, and Extra High. In other words, the model can process very long sets of documents while adjusting its calculation depth according to the task.
Google also positions it for code, analysis, and enterprise production-level workflows. Benchmarks published by the company place it high on several tests, with 3rd place on the Artificial Analysis Intelligence Index at 61, 2nd place on steerability in the Agent Arena Index, and top results on suites related to law, taxation, finance, or spreadsheets.
Pricing, availability, and a little game of musical chairs
The model is offered at $3 per million input tokens and $15 per million output tokens. Google states that Argon and Carbon temporarily cost $2 and $10 until November 30, before rising to $4 and $20, making Neon 25% cheaper in the long run in this comparison.
Neon is also available in Google Antigravity for paying AI Pro or Ultra subscribers. On the model picker side, Google revises the entire range with Auto for dynamic adaptation, Hydrogen for daily assistance, Neon for in-depth analysis, and Carbon for the most complex work. A hierarchy that has the merit of avoiding overly vague labels, even if it somewhat resembles a premium gas station menu.
By the numbers
- 1 million tokens — Neon’s context window, according to Google.
- 61 — Neon’s score on the Artificial Analysis Intelligence Index.
- $3 / 1 M tokens — Neon’s input price, according to Google.
- $15 / 1 M tokens — Neon’s output price, according to Google.
- November 30 — end of Argon and Carbon’s promotional pricing.