Photo: Tima Miroshnichenko / Pexels
The new model scores slightly higher than Gemini 4 Carbon while costing less per task, filling the gap between Google's flagship and its budget tier
Google DeepMind has a new model, and its main selling point is a smaller bill.
Gemini 4 Neon, unveiled on November 18, 2026, scores slightly higher than Gemini 4 Carbon. According to a widely shared post on X, it also costs about 17% less per task.
What Neon actually is #
Neon is the fourth model in the Gemini 4 series. It slots between two existing siblings with very different personalities.
Gemini 4 Carbon is the high-performance option, but it is also the slower one. Gemini 4 Hydrogen sits at the other end: faster and more economical. Neon is built to land in the middle, offering better speed and cost efficiency while keeping performance close to Carbon’s.
Google DeepMind is aiming the model at complex software engineering, knowledge work and scientific applications. These are tasks where a model gets called repeatedly, so small savings per call stack up quickly.
The numbers #
The cost picture is the headline act. The X post puts the savings at about 17% per task compared with Carbon.
At the default high thinking level, the savings look larger. There, Neon’s cost per task comes in approximately 32% below Carbon’s. “Thinking level” refers to how much reasoning the model does before answering. More thinking generally means better answers and a higher bill, so the setting you choose shapes the savings.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Raw pricing is set at $3 per million input tokens and $15 per million output tokens. Neon supports a 1-million-token context window.
On performance, Neon ranks #3 on the Artificial Analysis Intelligence Index, scoring 61 at its highest thinking level. It also takes the #2 spot on the Vals Index.
The model shows particular strength in legal research, finance, accounting and other analytical work. It also handles multi-file coding and tool execution, meaning it can work across several files and call outside software to complete a job.
Where you can use it #
Google did not keep Neon behind a waitlist. The model became available right after launch through three routes.
Developers can reach it through the Gemini API. Paid subscribers can use it in Google Antigravity. Regular users can find it in the Gemini app.
Background: a busy stretch for Gemini 4 #
Neon arrives during a rapid expansion of the Gemini 4 lineup in late 2026. Google DeepMind had already launched several models, including Argon and Carbon, before Neon appeared.
Neon targets high-frequency operational tasks. That category has become a pressure point as companies move AI out of pilot projects and into production systems, where the model runs constantly and every task carries a cost.
What this means #
For businesses already using Carbon, the math gets interesting. If Neon matches or slightly beats Carbon on quality while costing less, switching looks like an easy call for many workloads. The open question is how much the savings vary by thinking level, since the figures range from about 17% to approximately 32% depending on the comparison. Teams in legal, finance and accounting have the most obvious reason to test it. Those are the areas where Neon showed its strongest results, and they are also fields where firms process large volumes of documents.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.