FRIDAY, NOVEMBER 20, 2026 Tech news, minus the noise.
N News Cast
LATEST
Home  /  AI
AI

Google's Gemini 4 Neon Beats Carbon on Speed and Cost Per Task, and It Is Already Third in the World

Google released Gemini 4 Neon on November 18, a model that sits between the heavyweight Carbon and the quick Hydrogen, and it is already third on the independent Artificial Analysis index.

Google's Gemini 4 Neon Beats Carbon on Speed and Cost Per Task, and It Is Already Third in the World
Source: Google, public domain, via Wikimedia Commons

Google released Gemini 4 Neon on November 18, and the pitch is refreshingly specific. This is not another claim to be the smartest model ever built. Neon is the middle child of the Gemini 4 family on purpose: more capable than the fast Hydrogen, quicker and cheaper per finished task than the heavyweight Carbon. Google calls it frontier-level intelligence with fast end-to-end task completion, aimed at software engineering, knowledge work, and science. The early independent numbers suggest the positioning is honest, which is rarer than it should be in model launches.

Gemini 4 Neon: What the Benchmarks Actually Say

The independent scorekeepers at Artificial Analysis put Gemini 4 Neon at 61 on their Intelligence Index, third overall behind GPT-6.2 Astra and Claude Fable 5.2, and one point ahead of Gemini 4 Carbon. That single point matters less than the cost column next to it. Neon costs $6.44 per index task, about 17 percent less than Carbon, and finishes a task in roughly 21 minutes instead of 36, at about 110 output tokens per second. It is nearly twice as fast as its sibling and beats it on most benchmarks, according to The Decoder's read of the launch. For coding agents that run dozens of steps, that combination of speed and per-task price is the whole ballgame. A model that is slightly smarter but twice as slow costs you twice, in wall-clock time and in tokens burned on retries.

The pricing tells the same story from the other side. Neon costs $3 per million input tokens, $0.15 when cached, and $15 per million output tokens, with a 1-million-token context window and up to 64,000 tokens of default output. Per token, that is more than Carbon, which launched at $2 and $10. Per finished task, Neon is cheaper, because it gets there faster and wastes less. Anyone comparing models on sticker price alone will miss the point. What you pay for is completed work, and Neon completes it for less.

Where You Can Use Gemini 4 Neon

Neon is rolling out in the Gemini API in AI Studio, in Gemini for business, in Antigravity for paid Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra users, where it appears as "Deeper analysis" in the model menu and can also run Tasks. Google says Neon records its lowest fabrication rate yet in internal Antigravity testing, which is a claim worth watching rather than taking on faith, since fabrication rates are notoriously hard to compare across vendors. A cyber version of the model goes to vetted defenders in the coming days, following the pattern Google set with Gemini 4 Carbon, which reached security teams before the general public.

The practical question is who should switch. If you run coding agents or long agentic workflows, Neon is now the obvious default to test: it tops the coding-agent table in Antigravity, and the speed gain alone will show up in your iteration loop the same day. If you need raw reasoning power for the hardest problems, Carbon still has the edge on paper. If you need the cheapest possible responses at high volume, Gemini 4 Hydrogen still wins on price. Neon is for the middle, and the middle is where most real work lives.

Four Models in Seven Weeks

Neon is Google's fourth Gemini 4 model in seven weeks, after Hydrogen, Carbon, and Argon. That pace is the real story. Google has stopped shipping one flagship and hoping it covers everything. Instead it ships tiers, each with its own price and personality, and lets users pick the trade-off. OpenAI is doing the same thing: GPT-6.2 Astra and Sol arrived the same week, split across capability and cost in a similar way. The model market is turning into a menu, not a race with one winner.

The risk for Google is confusion. Four models with similar names, overlapping capabilities, and different prices is a lot to ask users to keep straight, and the naming does not help. Hydrogen, Carbon, Neon, Argon: chemistry-class branding that tells you nothing about what each model is for. The saving grace is that the differences are real and measurable. Neon is faster than Carbon and smarter than Hydrogen, and you can verify both claims yourself in an afternoon. In a year of benchmark claims that fall apart on contact with real work, that verifiability counts for something.

The Bottom Line on Neon

Gemini 4 Neon is not a revolution. It is a well-aimed middleweight, and that is exactly what the market needed from Google right now. It gives teams running agents a faster, cheaper-per-task option without asking them to give up frontier capability, and it sharpens the competition at the top of the index, where three labs are now separated by a handful of points. Try it on your longest, most annoying coding task first. That is where the speed and the price will either prove themselves or not, and you will know within an hour.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.