Google's Gemini 4 Neon: frontier brains at $3 a million tokens — and #1 for coding agents
Google launched Gemini 4 Neon on Wednesday, a frontier-level model priced at $3 per million input tokens that tops the Coding Agent Index — and undercuts its own flagship Carbon from December.
📊 Scorecard: our Bullish call on $GOOGL is a miss so far: from $282.44 to $272.66 at the latest close (-3.5%). Against QQQ the call is wrong: the stock is 1.3 points behind the index.

The News
Google launched Gemini 4 Neon on Wednesday, November 18, the fourth Gemini 4 model in seven weeks. It sits between the fast Hydrogen model and the heavy Carbon model, built for complex software engineering, knowledge work and science. It is priced at $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 — a 95% discount — a 1 million-token context window and four thinking levels that let developers trade speed for depth.
Google reports Neon at 61 on the Artificial Analysis Intelligence Index at its highest thinking level, third overall behind GPT-6.2 Astra and Claude Fable 5.2. It ranks second on the Vals Index at 70.9% and first on the Artificial Analysis Coding Agent Index at 58.6, with 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0. These are Google's launch-day figures and its citations of third-party leaderboards. Artificial Analysis independently tested Neon on launch day — scoring it 61 on the Intelligence Index and $6.44 per task — but no broader independent evaluation exists yet.
Neon is rolling out in the Gemini API (paid tier) and Google AI Studio, and in Antigravity and the Gemini app for paid Pro and Ultra subscribers. On price, Google claims Neon costs 32% less per task than Carbon at the default high setting, while Artificial Analysis independently measured about 17% lower — the same direction, a smaller gap. Carbon's introductory $2 per million input and $10 output pricing ends November 30 and rises to $4 and $20 on December 1, which would leave Neon cheaper than Carbon from December. Alphabet shares closed Wednesday at $282.44, up 0.80%.
Neon is Google's bet that the coding-agent race will be won on price as much as on brains.
Why It Matters
The launch matters because the coding-agent race is now fought on price as much as on capability. Developers and companies run millions of agent tasks a day, and a model that is nearly as capable as the flagship but meaningfully cheaper changes which model gets picked by default. Neon's number-one ranking on the Coding Agent Index gives Google a concrete claim to lead with, not just a price cut.
The pricing move is also a message to Google's own customers. Carbon's price is about to double from its introductory level, and Neon arrives just in time to catch anyone who balks at $4 per million input tokens. Google is effectively steering its heaviest users toward the cheaper model while keeping Carbon as the premium option — a tidy bit of product-line management, if the benchmarks hold up.
The honest caveat is the gap between Google's 32% cost claim and Artificial Analysis's 17% measurement. Both point the same way, but the difference is a reminder that launch-day benchmark numbers come from the seller. What matters now is whether independent testing over the next few weeks confirms Neon is genuinely close to Carbon on real coding work — and whether developers actually switch.
Analyst's Take
Our call is Bullish. A frontier-level model at $3 per million tokens, top of the coding-agent leaderboard, and cheaper than its own flagship from December is a strong hand for Google in the developer market. The stock's modest 0.80% rise on Wednesday suggests the market treated it as one more solid launch in a fast-moving sequence, not a turning point — and that may be the right read for now.
The risk is that benchmarks are not the same as adoption. GPT-6.2 Astra and Claude Fable 5.2 still sit ahead of Neon on the overall Intelligence Index, and the coding-agent crown is a narrow slice of the market. If developers try Neon and find the 17% saving is not worth switching from the tools they already use, the launch fades into the background of a very crowded model cycle.
What to watch next is usage: whether Neon shows up in developer surveys and API traffic in the coming weeks, and whether Google's December Carbon price rise pushes customers toward Neon as intended. Reader poll: Bullish or bearish on Alphabet after the Neon launch? Vote: Bullish, Bearish, Mixed.
Are you bullish or bearish on this story?
By The Numbers
FAQ
What is a token?
A token is a small chunk of text — roughly three-quarters of a word — that AI models read and write. When a model charges $3 per million input tokens, that is the price for feeding it about 750,000 words of text, and $15 per million output tokens is the price for the roughly 750,000 words it writes back.
What is a context window?
A context window is how much text a model can hold in mind at once. A 1 million-token window means Neon can work with roughly 750,000 words of code, documents or conversation in a single session without forgetting the start.
What is a coding agent benchmark?
It is a test that measures how well an AI model can do real software work on its own — writing code, fixing bugs and running commands — rather than just answering questions. The Coding Agent Index scores models on those tasks, so a #1 ranking means Neon did best on that test.
👁
Source: Google